Phosphoregulation of Cdt1 in G2 and M phases prevents re-replication independently of Geminin

DNA replication origin licensing, the process of MCM helicase loading, is an early essential step in replication and should be restricted to G1 phase to avoid re-replication and genome instability. Cdt1 is a critical MCM loading factor whose licensing activity must be restrained after G1. We discovered that Cdt1 hyperphosphorylation during G2 and M phase is essential to prevent re-replication and DNA damage, the first example of direct Cdt1 licensing activity control by post-translational modification. This hyperphosphorylation specifically requires Cyclin A/CDK1 and occurs at a cluster of phosphorylation sites in a disordered Cdt1 linker region. Hyperphosphorylation interferes with Cdt1-MCM binding independently of protein degradation or inhibition by the Cdt1 inhibitor, Geminin. At the M-G1 transition, Cdt1 is re-activated by protein phosphatase 1-dependent dephosphorylation. We propose that distinct, non-redundant re-replication inhibition mechanisms act in a sequential relay from early S through mitosis to ensure once, and only once, chromosome duplication.

"Cdt1-5A" (S391A, T402A, T406A, S411A, and S491A) is both virtually unphosphorylatable in vitro by stress-induced MAP kinases, and is compromised for G2 hyperphosphorylation detected by gel mobility shift 19 . Cdt1-5A bears alanine substitutions at five sites, and four are in a region of low sequence conservation and highpredicted intrinsic disorder ( Fig. 1a and Supplementary Fig. S1). This "linker" region connects the two domains of Cdt1 that have been structurally characterized for Geminin binding (a middle "M domain") and for MCM binding (C-terminal "C domain") 25,26 .
Because both cyclin dependent kinases (CDK) and MAP kinases share similar substrate sequence requirements (serine/threonine-proline), and because both are active in G2, we postulated that during normal G2 and M phases these Cdt1 sites are phosphorylated by CDK or MAPK 27 . We inserted cDNAs encoding either wild-type Cdt1 (Cdt1-WT) or Cdt1-5A into a single chromosomal FRT recombination site under doxycycline-inducible expression control in U2OS osteosarcoma cells. Both Cdt1 constructs bear C-terminal HA epitope and polyhistidine tags to distinguish ectopic Cdt1 from endogenous Cdt1.
As a measure of relative Cdt1 activity, we induced Cdt1 overproduction to approximately 5-10 times more than endogenous Cdt1 (Fig. 1b, compare lanes 1 and 2).
The amount of re-replication induced by Cdt1 overproduction is directly related to Cdt1 licensing activity 28 . As previously reported 29,30 , Cdt1-WT overproduction in human cells induced re-replication, which we detected by analytical flow cytometry as a population of cells with DNA content greater than the normal G2 amount (>4C, Fig. 1c, and Supplementary Figure S2b). Strikingly however, overproducing Cdt1-5A to the same level as Cdt1-WT induced substantially more re-replication suggesting that it is intrinsically more active. DNA re-replication can also induce the formation of giant nuclei 31, 32 , and we noted that the average nuclear area of cells overproducing Cdt1-WT was somewhat larger than control nuclei, whereas nuclei of cells overproducing Cdt1-5A were even larger ( Supplementary Fig. S3a). Thus, Cdt1-5A expression not only induces more cells to re-replicate, but it also induces a higher degree of re-replication in those cells than Cdt1-WT.
Re-replication is an aberrant genotoxic phenomenon characterized by molecular markers of DNA damage 6,33,34 . As an independent measure of re-replication, we analyzed lysates of Cdt1-overproducing cells for Chk1phosphorylation, a marker of the cellular DNA damage response. Cdt1-5A consistently induced more Chk1 phosphorylation than WT Cdt1 (Fig. 1d, compare lanes 2 and 3). We also noted that the accumulation of rereplicated cells came at the expense of G1 cells, consistent with a checkpoint arrest ( Supplementary Fig. S2b). Moreover, cells overproducing Cdt1-5A were also ~3 times more likely to generate -H2AX foci, another marker of re-replication-associated DNA damage 20,35,36 (Supplementary Fig. S3b). We thus conclude that phosphorylation at these sites negatively regulates Cdt1 activity.
In nocodazole-arrested (early mitotic) cells, phosphorylated Cdt1-WT migrated more slowly by SDS-PAGE than Cdt1-5A (Fig. 1e, lanes 2 and 5). As a better measure of Cdt1 phosphorylation, we analyzed Cdt1 gel migration in the presence of Phos-tag reagent which retards protein mobility proportional to the extent of phosphorylation 37 .
Endogenous Cdt1 from nocodazole-arrested cells migrates much more slowly on Phostag gels than endogenous Cdt1 from G1 cells, and this slow migration was reversed by phosphatase treatment (Supplementary Fig S2a). We detected similar slow Cdt1 migration in G2 cells synchronized by release from a thymidine arrest without the nocodazole block (data not shown), so we presume that the phosphorylation in prometaphase reflects phosphorylation from late S phase through mid-mitosis (i.e. "G2/M"). The distribution of ectopic Cdt1-5A bands was lower than Cdt1-WT bands on Phos-tag gels, demonstrating that these sites are indeed phosphorylated late in the cell cycle.
Phosphorylation at two additional candidate CDK/MAPK target sites in the linker region has been reported from global phosphoproteomics studies 38 . To test the potential additional contribution of these sites to Cdt1 regulation, we included the mutations S372A and S394A to Cdt1-5A to create Cdt1-7A (Fig. 1a). This variant had slightly increased mobility on Phos-tag gels relative to Cdt1-5A (Fig. 1e, compare lanes 5 and 6).
From this observation, we infer that Cdt1-5A is already at the lowest activity that can be achieved from phosphorylation in the linker region, and that additional phosphorylations do not cause further activity decreases.
Cdt1 is also known to be phosphorylated at both T29 and S31 38,39 . Phosphorylation at T29 generates a binding site for the SCF Skp2 E3 ubiquitin ligase, which contributes to Cdt1 degradation during S phase 40,41 . Robust Cdt1 degradation in S phase is important for avoiding re-replication 30,42,43 . The stress MAPK JNK (c-Jun N-terminal kinase) has also been reported to inhibit Cdt1 by phosphorylating T29 39 . To determine if these N-terminal phosphorylations add to the effects of linker region phosphorylations, we added the two mutations, T29A and S31A, to Cdt1-7A to generate Cdt1-9A. Cdt1-9A from nocodazolearrested cells migrated even faster than Cdt1-7A on Phos-tag gels (similar to phosphatase-treated WT Cdt1, not shown), demonstrating that one or both T29 and S31 are phosphorylated after S phase, although Cdt1 is not as unstable in G2 as it is in S phase. Cdt1-9A overproduction induced even more re-replication than Cdt1 bearing only linker region mutations, Cdt1-4A, 5A, and 7A (Fig. 1c), and similar amounts of DNA damage checkpoint activation (pChk1, Fig 1d, lanes 5 and 9). We presume that compromised S phase degradation from loss of SCF Skp2 targeting contributes to this enhanced re-replication 30,41,44 .
Cyclin A/CDK1 is the primary Cdt1 kinase during G2 and M phases.
To determine which kinase(s) is responsible for Cdt1 inactivation, we assessed the effects of kinase inhibitors. All nine of the sites in Cdt1-9A can be targeted by both CDKs and MAPKs since all nine are serine or threonine followed by proline ( Supplementary Fig. S1). We synchronized cells in nocodazole when Cdt1 is maximally phosphorylated and then tested the effects of pharmacological MAPK and CDK inhibitors on the migration of endogenous Cdt1 by Phos-tag gel analysis. We first treated nocodazole-arrested cells with pharmacological inhibitors of p38 or JNK, two stressactivated MAP kinases which we previously showed can phosphorylate the linker region and inactivate origin licensing during a stress response 19 (p38 inhibitor SB203580 and c-Jun N-terminal kinase JNK inhibitor VIII). These MAPK inhibitors, either alone or in combination, had no effect on mitotic Cdt1 migration on Phos-tag gels (Fig. 2a, lanes 8-10, compared to lane 4). We confirmed that the inhibitors were active in these cells at these concentrations by analyzing known downstream substrates (Supplementary Fig.   S4a-c) 19,[45][46][47][48] . We also tested inhibitors of CDK1 and CDK2 singly or in combination.
The slow migration of phospho-Cdt1 was largely reversed by treatment with CDK1 inhibitor RO3306 for just 15 minutes (Fig. 2a, compare lanes 5, 7, and 11 to lane 4), but not when treated with the CDK2 inhibitor CVT313, (Fig. 2a, lane 6). This effect occurred even in the presence of the proteasome inhibitor MG132, which we included in all kinase inhibitor experiments to prevent premature anaphase onset. CDK1 is normally activated by either Cyclin A or Cyclin B, and we sought to determine which cyclin is responsible for directing CDK1 to phosphorylate Cdt1. We therefore took advantage of the polyhistidine tag at the C-terminus of the Cdt1-WT construct to retrieve Cdt1 from lysates of transiently transfected, nocodazole-arrested 293T cells. As a control, we included a Cdt1 variant with a previously-characterized mutation in the cyclin binding motif, Cdt1-Cy 41 (RRL to AAA at positions 66-68, Fig.   1a). We analyzed His-Cdt1-bound proteins from these lysates for the presence of endogenous cyclin and CDK subunits. Cdt1-WT interacted with both CDK1 and CDK2, and strongly interacted with Cyclin A, but not at all with either Cyclin B or Cyclin E ( Fig. 2b). Cdt1-Cy retrieved no cyclins or CDKs, indicating that the only CDK binding site in Cdt1 is the RRL at positions 66-68. Since Cdt1 binds Cyclin A, CDK1 and CDK2, but only inhibition of CDK1 activity affected Cdt1 phosphorylation, we conclude that Cyclin A/CDK1 is responsible for the inactivating Cdt1 phosphorylations during G2 and M phases.

Cdt1 phosphorylation blocks MCM binding.
We next sought to determine by what mechanism Cyclin A/CDK1-mediated phosphorylation inhibits Cdt1 licensing activity. The inhibitory phosphorylation sites are in the linker region between the middle and C-terminal domains (Fig. 1a), and these positions are not visible in any currently available Cdt1 atomic structures. Nonetheless, our recently-generated homology model of the human Cdt1-MCM complex 49 led us to speculate that phosphorylation-induced changes at this linker could inhibit MCM binding, either because the linker contacts MCM directly or because phosphorylation alters Cdt1 conformation or relative positions of the two MCM binding domains (Fig.   3a). We thus set out to test if MCM interacts with hypophosphorylated G1 Cdt1 more effectively than with hyperphosphorylated G2 Cdt1 (i.e. if phosphorylation impairs Cdt1-MCM binding). We noted however that simply comparing co-immunoprecipitations from lysates of G1 and G2 phase cells is complicated by the presence of the Cdt1 inhibitor, Geminin, which interferes with the Cdt1-MCM interaction 50,51 and is only present in the G2 cells. Because Geminin is differentially expressed in G1 and G2 cells, the comparison would not be fair. To account for the effects of Geminin, we prepared a lysate of asynchronously-proliferating cells which contain mostly G1 hypophosphorylated Cdt1; Cdt1 is degraded in S phase, and cells spend a relatively small fraction of total cell cycle time in G2 (e.g. Supplementary Fig. S2b). We mixed this lysate with lysate from nocodazole-arrested cells that contains both Geminin and hyperphosphorylated Cdt1. In this way, we created a similar opportunity for MCM to bind either hyper-or hypophosphorylated Cdt1. We then immunoprecipitated endogenous MCM2 and probed for MCM6 as a marker of the MCM complex and for tagged Cdt1. As a control, we immunoprecipitated MCM2 from an unmixed lysate of nocodazole-arrested cells. As expected, Geminin did not co-precipitate with MCM since the Cdt1-Geminin and Cdt1-MCM interactions are mutually exclusive 50,51 (Fig. 3b). We found that the MCM complex retrieved from the mixed lysates was enriched for the faster-migrating hypophosphorylated Cdt1 relative to hyperphosphorylated Cdt1 and that the total amount of Cdt1 bound to MCM was much higher when hypophosphorylated Cdt1 was available  4). In summary, these results suggest that Cdt1 phosphorylation disrupts its interaction with MCM complex, and that this disruption contributes to re-replication inhibition in G2 and M phases. We note that this is the first example of direct regulation of the Cdt1-MCM interaction by post-translational modification.

Cdt1 dephosphorylation at the M-G1 transition requires PP1 phosphatase activity.
Our finding that Cdt1 phosphorylation in G2 and M phase inhibits its ability to bind MCM suggests that Cdt1 must be dephosphorylated in the subsequent G1 phase to restore its normal function. To explore this notion, we first monitored Cdt1 expression and phosphorylation in cells progressing from M phase into G1. We released nocodazolearrested cells and collected time points for analysis by immunoblotting (Fig. 4a).
Geminin is a substrate of the Anaphase Promoting Complex/Cyclosome (APC/C) 52 , and as expected for an APC/C substrate, Geminin was rapidly degraded within 60 minutes of mitotic release. In contrast, Cdt1 was not degraded during the M-G1 transition but rather, was rapidly dephosphorylated coincident with Geminin degradation (Fig. 4a, compare lanes 3 and 4). We next explored which phosphatase is required for Cdt1 dephosphorylation. We first tested phosphatase inhibitors for the ability to prevent Cdt1 dephosphorylation after CDK1 inhibition. We tested inhibitors of protein phosphatase 1 (PP1) and protein phosphatase 2A (PP2A), which are responsible for the majority of protein dephosphorylation in cells 53 . We treated nocodazole-arrested cells with the CDK1 inhibitor to induce Cdt1 dephosphorylation in the presence or absence of calyculin A (Cal A) or okadaic acid (OA) 54 . Both compounds are potent inhibitors of both PP1 and PP2A, but calyculin A is more effective than okadaic acid for inhibiting PP1, particularly at the concentrations we tested 55,56 . We found that calyculin A maintained Cdt1 hyperphosphorylation (Fig. 4b, compare lanes 2 and 3) whereas concentrations of okadaic acid that inhibit PP2A but not PP1 did not affect Cdt1 dephosphorylation ( Supplementary Fig. S5). In addition, we released nocodazole-arrested cells into G1 phase for 30 minutes (to initiate mitotic progression) and then treated the cells with calyculin A. As a control, we probed for MCM4, a known PP1 substrate that is normally dephosphorylated in G1 phase 57 ; calyculin A largely prevented MCM4 dephosphorylation (Fig. 4c). PP1 inhibition also largely prevented Cdt1 dephosphorylation during the mitosis-G1 phase transition without blocking mitotic progression as evidenced by Geminin degradation (Fig. 4c, lanes 2 and 3). These results suggest that the PP1 family phosphatase is required for Cdt1 dephosphorylation. By extension, we suggest that PP1 activity is required to re-activate Cdt1-MCM binding and origin licensing in G1 phase.

Cell cycle-dependent Cdt1 phosphorylation
Mammalian Cdt1 is degraded during S phase, and this degradation is essential to prevent re-replication 18,30,42, . Perhaps counter-intuitively, Cdt1 then accumulates beginning in late S phase 58 , and by mitosis reaches a level similar to Cdt1 in G1 phase 59,60 . Despite the potential danger from re-licensing and re-replicating G2 DNA, these high Cdt1 levels serve two purposes: 1) Cdt1 is essential for stable kinetochore-microtubule attachments 61 , and 2) high levels of Cdt1 in mitosis can improve licensing efficiency in the next G1 phase 60 . In this study, we discovered that Cdt1 phosphorylation by Cyclin A/CDK1, a kinase active during G2 phase, inhibits Cdt1 licensing activity and contributes to preventing DNA re-replication while Cdt1 levels are high in G2 and M phase.
We analyzed a cluster of inhibitory Cyclin A/CDK1 phosphorylation sites that are distinct from the only previously-characterized CDK sites at T29 and S31. T29 phosphorylation contributes to Cdt1 degradation during S phase by creating a binding site for the SCF Skp2 E3 ubiquitin ligase. The sites we identified here, S391, T402, T406, S411, and S491, do not induce Cdt1 degradation however. In fact, we previously demonstrated that phosphorylation at these sites stabilizes, rather than destabilizes Cdt1 19 . Aside from separating the four linker sites from the most C-terminal S491 site in the 5A allele, we did not attempt to systematically dissect these "linker" phosphorylation sites, largely because this region of Cdt1 is not strongly conserved among vertebrates ( Fig. 1a and Supplementary Fig. S1). Most vertebrate Cdt1 linker sequences are nonetheless predicted to be similarly disordered, and most have at least one candidate CDK phosphorylation site ( Supplementary Fig. S1). Interestingly, altering two additional sites in this region (converting Cdt1-5A to Cdt1-7A) did not exacerbate the re-replication phenotype suggesting that four phosphorylations are sufficient to achieve maximal Cdt1 inhibition.
It may be that the four linker sites vary in their relative importance for inhibiting human Cdt1 activity, or it may simply be the need for a total amount of phosphorylation in this region regardless of specific position. In that regard, multisite Cdt1 linker phosphorylation may resemble other examples of cell cycle-dependent multisite phosphorylation in which the total negative charge is more important than the specific phosphorylated position 62 . If so, then all four sites in Cdt1 may work additively to achieve maximal inhibition.
Although we had previously established that stress-activated MAP kinases (p38 and JNK) can phosphorylate these inhibitory sites in Cdt1 during a stress response, and both p38 and JNK are active during a G2 arrest, we could detect no contribution of stress MAPK activity to endogenous Cdt1 phosphorylation during G2 and M phases. (The ability of a JNK inhibitor to reverse Cdt1 phosphorylation in nocodazole-arrested cells may be attributed to off-target indirect effects of the drug on CDK1 activity 63 (Fig. 3a). Of additional note the linker phosphorylations are not in the Cdt1 domain that is both necessary and sufficient for Geminin binding 25,75 . Thus as expected and consistent with previous findings 40 , this mutation does not affect the binding of Cdt1 to Geminin (data not shown), indicating that the contribution of the phosphorylation site mutations to induced re-replication is Geminin-independent.

PP1-dependent Cdt1 dephosphorylation
Approximately one-third of all eukaryotic proteins may be dephosphorylated by PP1 53 . PP1 binds some of its substrates directly via a short motif, RVxF, KGILK or RKLHY 53,76 . Human Cdt1 contains several such predicted PP1 binding motifs and thus may be a direct target of PP1. Alternatively, Cdt1 dephosphorylation may require an adapter to bind PP1 similar to the role of the Rif1 adapter for MCM dephosphorylation 77 .
In either case, the fact that hyperphosphorylated Cdt1 binds MCM poorly, plus the fact that the levels of Cdt1 do not change from M phase to G1 (i.e. Cdt1 is not degraded and resynthesized at the M-G1 transition), means that PP1-dependent Cdt1 dephosphorylation activates origin licensing. In that regard, dephosphorylation is the first example of direct Cdt1 activation, and it complements indirect activation by Geminin degradation in M phase.

A sequential relay of re-replication inhibition mechanisms
We propose that Cdt1 activity is restricted to G1 through multiple regulatory mechanisms during a single cell cycle, but that the relative importance of individual mechanisms changes at different times after G1 (Fig. 5). At the onset of S phase Cdt1 is first subjected to rapid replication-coupled destruction via CRL4 Cdt2 which targets Cdt1 bound to DNA-loaded PCNA 78 . This degradation alone is not sufficient to prevent rereplication however, and a contribution from Cyclin A/CDK2 to create a binding site for the SCF Skp2 E3 ubiquitin ligase is also essential 30 . We suggest that SCF Skp2 -targeting occurs primarily in mid and late S phase based on the dynamics of Cyclin A accumulation 41,79 . A reinforcing mechanism for Cdt1 degradation is more important in mid and late S phase than in early S phase because the amount of DNA that has already been copied increases throughout S phase. Licensing DNA that hasn't been copied yet is presumably benign, but as S phase proceeds, the amount of DNA that has been copied already (i.e. the substrate for re-replication) also increases. The Cdt1 inhibitor, Geminin, begins to accumulate in early S phase, and its levels increase along with the amount of replicated DNA until Geminin is targeted for degradation by the APC/C during mitosis 21,52 . Geminin binding to Cdt1 interferes with Cdt1-MCM binding, and since Cdt1-MCM binding is essential for MCM loading, Geminin prevents re-licensing 31,32 .
This inhibition is particularly important once Cdt1 re-accumulates after S phase is complete; in late S phase the responsibility for restraining Cdt1 is passed from the ubiquitin ligases to Geminin and Cyclin A/CDK1. Just as CRL4 Cdt2 -mediated degradation in S phase is not sufficient to fully prevent re-replication, we demonstrated that the presence of Geminin alone is not sufficient to inhibit Cdt1 during G2. Cyclin A/CDK1mediated Cdt1 phosphorylation in a linker domain between two MCM binding sites 49,72,73,74 also prevents Cdt1-MCM binding. These (and potentially more) mechanisms to restrain Cdt1 activity are also reinforced by regulation to inhibit ORC, Cdc6, PR-Set7, and other licensing activators 3,16,29,80 . Given that there are many thousands of potential origins in mammalian genomes, and the consequences of even a small amount of rereplication are potentially dire, precise once-and-only-once replication requires that Cdt1 be inhibited by at least two mechanisms at all times from G1 through mitosis.

Sequence analysis.
A representative selection of vertebrate sequences for comparison was taken from Miller et al. 81

Statistical analysis
The differences were considered significant with a p-value less than 0.05 in an unpaired Student t-test or a Mann-Whitney U -test using Sigmaplot software (Systat Software).         The phosphorylation and total protein levels of MK2 were analyzed by immunoblotting.