Evolution of opposing regulatory interactions underlies the emergence of eukaryotic cell cycle checkpoints

In eukaryotes the entry into mitosis is initiated by activation of cyclin-dependent kinases (CDKs), which in turn activate a large number of protein kinases to induce all mitotic processes. The general view is that kinases are active in mitosis and phosphatases turn them off in interphase. Kinases activate each other by cross- and self-phosphorylation, while phosphatases remove these phosphate groups to inactivate kinases. Crucial exceptions to this general rule are the interphase kinase Wee1 and the mitotic phosphatase Cdc25. Together they directly control CDK in an opposite way of the general rule of mitotic phosphorylation and interphase dephosphorylation. Here we investigate why this opposite system emerged and got fixed in almost all eukaryotes. Our results show that this reversed action of a kinase-phosphatase pair, Wee1 and Cdc25, on CDK is particularly suited to establish a stable G2 phase and to add checkpoints to the cell cycle. We show that all these regulators appeared together in LECA (Last Eukaryote Common Ancestor) and co-evolved in eukaryotes, suggesting that this twist in kinase-phosphatase regulation was a crucial step happening at the emergence of eukaryotes.


Results
The dynamic behaviour of the mitotic regulatory network can be emulated by a simpler network 36 , where a single kinase and phosphatase interact (Fig. 1). In this Mutual Inhibition (MI) system the kinase activates itself and inactivates the phosphatase by phosphorylation, while the phosphatase removes these phosphate groups, thereby activating itself and inhibiting the kinase (Fig. 1a).
Mitotic entry can be emulated by the MI system by setting the reaction rates of all interactions equal to 1 (Supplementary Table S1) 36 . Figure 1 shows how the regulatory components of the mitotic system could be grouped based on their dynamical properties. Pho of the MI system ( Fig. 1a and Supplementary Figure S2) groups the phosphatases PP2A and PP1 together with the kinase Wee1 ( Fig. 1b and Supplementary Figure S3). The dynamical behaviour of these three molecular species overlap with the behaviour of the molecular species Pho of the MI system (Fig. 1c,d). Similarly, Kin in the MI system groups the dynamics of the kinases Cdk1 and Gwl together with the phosphatase Cdc25 (Fig. 1). These two groups of species from the mitotic entry system correspond with the biological functions of a 'kinase' that promotes mitosis and a 'phosphatase' that promotes interphase, but the biochemical types of Wee1 and Cdc25 are opposite to their corresponding biological functions. The simplified network of mitotic mutual inhibition (MI; Fig. 1a) emphasizes the tug of war between kinases and phosphatases at the transition from phosphatase-dominated interphase to kinase-dominated mitosis. The general framework is that mitotic kinases (respectively, interphase phosphatases) activate each other and inhibit interphase phosphatases (respectively, mitotic kinases), but Cdc25 and Wee1 clearly break this generalized rule.
Cdc25 is not a typical mitotic phosphatase; it has evolved from the rhodanese-like family of phosphatases, far distinct from PP1 and PP2A 17,19,39 . Based on its sequence and structure, Wee1 is classified together with Gwl and Cdk1 in the serine/threonine protein kinase family; however, Wee1 functions as a tyrosine-specific protein kinase 40 . Furthermore, Wee1 and Cdc25 are distinct from other members of the regulatory network of mitotic onset. Typical kinases, such as Gwl and Cdk1, must be phosphorylated on an activation domain to adopt an active conformation 6 . Conversely, typical phosphatases, such as PP1 and PP2A, are active in their dephosphorylated conformation 7,41 . The regulation of Wee1 and Cdc25 does not align with the regulation of these canonical kinases  Figure S1). Shadows of molecules in the G2M system point to their corresponding species in the MI system. (c) Time-course diagram of MI system. (d) Time-course diagram of the G2/M system. Diagrams show the active forms from each species. Active forms of Cdk1, Cdc25 and Gwl collapse into the 'Mitosis' trace of MI (brown), while active forms of PP2A, PP1 and Wee1 collapse into the 'Interphase' trace of MI (grey). See supplementary methods for details of the simulations. Graphics were obtained by R version 3.6 and ggplot2 37,38 . and phosphatases. Contrary to Cdk1 and Gwl, Wee1 does not need to be phosphorylated for activation 40 ; rather, Wee1 is inhibited by phosphorylation by Cdk1. Furthermore, Cdc25 requires phosphorylation for its activation, in contrast to other phosphatases, PP2A and PP1, which are active when they are dephosphorylated 42 . The atypical activations of Wee1 and Cdc25 are also reflected in their regulation of Cdk1. Wee1 phosphorylates Cdk1 at an inhibitory site, and the active form of Cdk1 is recovered by dephosphorylation by the phosphatase Cdc25 (Note: Cdk1 has the usual activatory phosphorylation site, phosphorylated by CAK 43 , and Wee1 activity can be increased by phosphorylation on sites other than the Cdk1 target 44 , but these phosphorylations are not involved in the feedback loops we investigate here).
In conclusion, we observe that the biological functions of Wee1 and Cdc25 do not agree with their biochemical functions, and the standard roles of phosphatases and kinases is upside-down in the Cdk1/Wee1/Cdc25 core. It is yet unclear why this atypical regulation of kinases and phosphatases can be observed in the control of a crucial cell cycle transition. In the next section, we investigate the impact of this twist in functions on the dynamical behaviour of the entry into mitosis. Specifically, we compare the dynamical properties of the mitotic entry network with alternative versions of the control system, where the biochemical and biological functions of the regulatory proteins are consistent. Figure 1 showed that the biochemical functions of Wee1 and Cdc25 do not match with the mitotic role of typical kinases and phosphatases respectively. To understand the implications of this rewiring on the dynamical behaviour of the mitotic onset, we introduce a model of three kinases and three phosphatases, we term as the General Kinase-Phosphatase (GKP) system ( Fig. 2a and Supplementary Figure S4). This system is composed of three copies of interlocked Mutual Inhibition (MI) systems by multiplying components while keeping their biochemical and biological functions in agreement. We are investigating how far the dynamical behaviour of this system differs from that of the real mitotic entry regulatory system with its atypical kinases and phosphatase.

Cell cycle regulation with atypical kinases and phosphatases
The GKP system shows the same topology as the mitotic entry network (Fig. 2b). However, the kinase Wee1 is replaced by a phosphatase that is inactivating Cdk1, maintaining the coupling between biochemical and biological functions (Pho2 in Fig. 2a). Similarly, the phosphatase Cdc25 is substituted by a kinase and groups with the mitotic factors (Kin1 in Fig. 2a). The other components of the GKP system work exactly as the molecular species of the mitotic entry network (G2/M system, Fig. 2b). Since the GKP system does not exist in nature, we have replaced all the real names by the abbreviation of their biochemical types.
Solely based on the topology of the G2/M network and the GKP system, there should not be a real difference between their dynamical behaviours ( Fig. 2a,b). However, we observe a major discrepancy in the way kinases and phosphatases control the (de)phosphorylation cycles 6,7,46 . The function of kinases is tightly linked to the presence of ATP, as these molecules serve as both energy and phosphate source [47][48][49][50] . Phosphatases do not demand an energy source for dephosphorylation and use H 2 O molecules as phosphate acceptors to transfer the phosphate group from the phosphoprotein 7,46 .
The cell cycle is a highly energetic demanding process, whose progression is tightly connected with metabolism [51][52][53] . The oscillations of ATP concentration were proposed to be connected to the cell cycle progression 48,[53][54][55][56] . In the G1/S phase it is found in its minima; it reaches its peak in late G2 phase, and rapidly drops during mitosis 51,55,57 The ATP/ADP ratio recently re-emerged as a dynamical driver of the cell cycle, since it is directly controlled by redox reactions and pH 51,53,58-60 . The ATP/ADP ratio clearly controls the rates of phosphorylation reactions [48][49][50] , but there is less evidence about the direct proportionality of reaction rates to ATP/ADP ratio 61,62 . Furthermore, it has been recently proposed that the switch in the activities of the Cdk1-Cdc25-Wee1 core module can be driven by changes in the ATP/ADP ratio of cells 63 .
Following these lines, we investigate the steady states of both systems (G2/M and GKP) at various levels of ATP/ADP ratio (Fig. 2). The wiring diagrams of Fig. 2a,b show which reactions of the GKP system and the mitotic entry system rely on ATP. For both systems we can see the typical S-shape bifurcation plots of bistable systems 64,65 , since in a given ATP/ADP ratio regime the systems can settle both in a high and a low Kin0 or Cdk1 (Fig. 2c,d respectively) activity states. Similar pictures were drawn for almost all cell cycle transitions, mostly using cell size or abundance of cyclins as external drivers of the system 5,64,66,67 . Here we see that the G2/M system we consider can be driven to mitosis only at a high ATP/ADP ratio (7.77), when the low Cdk1 activity steady state disappears. On the other hand, the GKP system can keep the corresponding Kin0 at a low steady state only up to an ATP/ ADP ratio of 1.53 (Fig. 2c,d). This sixfold change difference between the two models is maintained and even more prominent as we change the reaction rates of the two systems parallelly (Fig. 2e). Here we plotted how the bistable region for ATP/ADP ratio changes widens as we increase k 1 , the parameter that is controlling the strength of all the enzymatic reactions. When k 1 is small, then background, slow phosphorylation and dephosphorylation reactions maintain a single steady state, but as it increases, the positive feedbacks take over and bistability appears.
The modelled systems of Fig. 2 are certainly simplified versions of the real system. We used equal reaction rates and basic two step modification kinetics in each reaction to simulate a generic model, not specifically fitting to any organism where this network is present (see supplement for details). However, the qualitative characteristics proposed by this analysis, highlights the central role of flipping Wee1 kinase and Cdc25 phosphatase in the G2/M system. The flip could have been essential to create a control on the initiation of mitosis when cells were in an energy rich environment and loaded with ATP. Without this flip, the GKP model cannot provide a wide and stable low kinase activity state, even at low ATP/ADP ratios, below what is naturally observed in eukaryotes 53

Controlling the entry into mitosis
Mitotic entry requires not only the presence of enough ATP in the cells, critical checkpoints ensure that mitosis should start only if the DNA is in a proper condition for mitosis. Any DNA damage needs to be repaired in the G2 phase before the cells enter into mitosis, so cells need to be able to stop mitotic entry upon such damages [69][70][71] . G2 arrest upon DNA damage is achieved through the activation of the widely conserved ATM-dependent checkpoint kinases (Chk1 and Chk2) 69,70 . When DNA damage is detected, ATM phosphorylates and activates both Chk1 and Chk2 and in their active form, the checkpoint kinases phosphorylate both Wee1 and Cdc25 [71][72][73][74] . In this case, the phosphorylation of the kinase Wee1 leads to an increase in its activity (or reduced degradation), and the phosphorylation of the phosphatase Cdc25 reduces its activity (or induces its removal). Simplified forms of these interactions were incorporated in a model presented on Fig. 3a and detailed on d Supplementary Figure S5. This system cannot be driven into mitosis by high ATP levels when a small amount of Chk is present (Fig. 3b), still a minimal amount of ATP is needed to bring Cdk1 to 0 level, since Chk also requires some ATP. On Fig. 3b we plot how the level of Chk affects the steady states of the Cdk1 module in the presence of high ATP levels. It is clearly observable that Cdk1 and Cdc25 activities go down to zero and Wee1 activity to maximum already at a  Remarkably, the checkpoint kinases regulate Wee1 and Cdc25 as if they were typical kinase and phosphatase [72][73][74] . The phosphorylation event activates the kinase Wee1 and inhibits the phosphatase Cdc25. These interactions do not add any extra twists, keeping the Wee1-Cdc25-Cdk1 core as the unique atypical regulation inside the regulatory network of the mitotic entry and possibly suggesting that the regulation of Wee1 and Cdc25 by Chk might be more ancient than the twisted effects of Cdk1. It is worth noting that the Wee1-Cdc25 network is not essential for eukaryotes, but cells missing these molecules are more sensitive to DNA damage 75,76 .
This efficient and simple mechanism of arresting the G2/M transition seems to rely on the upside-down regulation of the Cdk1-Cdc25-Wee1 core. To investigate the effect of this atypical regulation in the blocking of mitotic entry, the GKP system is investigated under the same control by a checkpoint (Fig. 3). In the GKP system phosphatases keep the cells in interphase and kinases drive mitosis. To preserve this property, the checkpoint is controlled by a checkpoint phosphatase (Chp on Fig. 3c and Supplementary Figure S6). This phosphatase Chp controls Kin0 (Cdk1) through the activation of the inhibitor Pho2 (Wee1), and inhibition of the activator Kin1 (Cdc25).
The GKP system affected by the checkpoint phosphatase Chp shows a bistable response both for changes in ATP and Chp levels (Fig. 3). The phosphatase active state can be stabilised only up to a given ATP/ADP ratio, above which kinases eventually overtake and win (Fig. 3d). The increase in Chp level can lead to a stable state with kinases losing, but a kinase winning steady state could still exist until large Chp levels (Fig. 3d). Thus, Chp can increase the threshold where kinases can take over but cannot stabilise the phosphatase winning state for high ATP/ADP ratios. Similar to the checkpoint kinase Chk in the mitotic system (Fig. 3a), the checkpoint phosphatase Chp counteracts the antagonistic feedback loop between Kin0 (Cdk1) and its inhibitor Pho2 (Wee1), and the pure positive loop between Kin0 (Cdk1) and its activator Kin1 (Cdc25) (Fig. 3c). However, high ATP levels can overtake this effect in the GKP system, while in the real mitotic entry network ATP alone cannot drive cells with DNA damage into mitosis. In summary, a system such as the GKP, where all kinases and phosphatases are regulated according to their biochemical function, cannot block the entry into mitosis. Presumably evolution would select against this type of systems, when cells are in an environment highly enriched in energy sources. In contrast, the widely observed atypical regulation of the Cdk1-Cdc25-Wee1 core can induce a stable arrest in G2 phase, which would be selected for.

Evolutionary perspective on the role of the kinase-phosphatase switch in the emergence of eukaryotes
In the previous sections, we have shown that the twisted regulation of Wee1 and Cdc25 plays a major role in the DNA damage-controlled entry into mitosis. They are not only regulated in the opposite way than the rest of kinases and phosphatases, but they also control CDK in an unorthodox way. The topology of the regulatory network, together with this upside-down regulation, creates an efficient and simple mechanism to arrest the cell cycle in the G2 phase. Here we investigate how this network emerged during the evolution of eukaryotes and evolved together thereafter. The commitment into mitosis is a highly energy demanding process. During the G2 phase, ATP is accumulated to be used later in mitosis 55 . Early events of the mitotic onset, such as chromosome condensation or spindle assembly 1,8 need a substantial amount of ATP. However, the events that take place later in anaphase are the ones that consume most of the stored energy 55 . Thus, a system to keep track of the available energy evolved early during evolution 77 . In eukaryotic organisms, this relies on the AMP-activated protein kinase (AMPK) [78][79][80] , which controls several energy dependent processes, including cell cycle progression 79 . Early eukaryotic cell cycle control systems might have evolved in a way that they allowed cells to enter into mitosis only if cellular energy level reached a critical threshold 81 . Another, an even more ancient system, where energy demand can drive oscillations is the cyanobacterial circadian clock [82][83][84][85] . Processes that require high levels of energy are active over the day phase, and the low energy processes occur during the night phase. In many organisms the cell cycle and the circadian clock are coupled 86 , and DNA damage is controlling and further coupling both of them 87 , so there is a good chance that the circadian clock played a crucial role in ancient metabolic control of cell cycle regulation. Apart from timekeepers, primitive biological systems may also hold a mechanism for the control of their cell division. We have shown before 81 that an antagonistic system of a single kinase-phosphatase pair (MI) is enough to sense energy level and allow kinase activation only above a critical level. In this framework, it is feasible to presume that early kinases may acquire the function of mitotic factors. When the energy source piled up, the kinases increased their activity, promoting the division through the phosphorylation of their substrates.
The ancient network that regulated cell division in the FECA (first eukaryotic common ancestor), certainly did not contain all factors involved in the current mitotic system. However, some of the key cell division regulators emerged before the appearance of LECA (last eukaryotic common ancestor 88,89 ). Since primitive kinases took the function of mitotic factors, molecules that counteract their effect could have taken the opposite role. These last ones could have been the heirs of current phosphatases. Thus, just a primitive group of kinases and phosphatases may have controlled the earliest processes of eukaryotic cell division. Through phylogenetic analyses of correlated evolution, we have investigated Wee1, Cdc25, Cdk1, Chk1 and Chk2 genes (Fig. 4a,b, see also Supplementary  Table 2). Ancestral state reconstructions suggest that Wee1, Cdc25, Cdk1 and Chk2 were already present in LECA, and Chk1 appeared in the Amorphea (which includes Opisthokonta and Evosea). That is, of the two checkpoint kinases, Chk2 might be ancient and Chk1 emerged much later afterwards, although it should be noted that denser sampling of early eukaryotes might provide a higher resolution view on the sequence of emergence of these genes. They remained conserved in most eukaryotes ( Supplementary Fig. S2), except in the lineage leading to Archaeplastida (the group including plants), where we inferred losses of Cdc25 and Chk2 ( Supplementary Fig. S7). It has been postulated that the function of Cdc25 was replaced by CDKs and B-type cyclins 70,71 and that DNA damage checkpoints use an alternative pathway to stop the cell cycle in plants 87,90 . Indeed, the loss of these genes could be responsible for an increased sensitivity of plants to DNA damage 91 . We used the likelihood ratio test (LRT) based on Pagel 92 and Barker et al. 93 to investigate correlated evolution between the genes above. We inferred that the cell cycle regulator Cdk1 show correlated evolution with Wee1 (p ≤ 0.05, LRT), Cdc25 (p ≤ 0.05, LRT) and Chk2 (p ≤ 0.001, LRT), and that Wee1 show correlated evolution with Chk1 (p ≤ 0.05, LRT) (Fig. 4b). We also investigated the co-evolution of these cell cycle G2/M transition regulators with Cks1 and NDR. Cks1 has principal roles in cell cycle regulation as an essential, highly conserved binding partner of CDKs 94 . NDR is a conserved Nuclear Dbf2-Related kinase, which carries out an essential function in late mitosis 95 . Since Cks1 is always present in complex with CDK 96 , this can serve as a positive control for proteins co-evolving with Cdk1. NDR is also a conserved protein kinase, but not in direct connection with CDKs and a role in G2/M transition, so this can serve as a negative control, as a cell cycle kinase that is not expected to closely co-evolve with Cdk1. Our analysis shows that four of the examined proteins (Cdc25, Cdk1, Chk1 and Chk2) showed a significant correlation with Cks1 and only Wee1 showed a correlation with NDR (Fig. 4c).
These analyses show that the Cdk1-Wee1-Cdc25-Chk2 network appeared and co-evolved together to provide a stable G2/M checkpoint for eukaryotes. This finding further supports the claim that the twist in kinasephosphatase activities at the G2/M transition regulation was a key evolutionary step to create a checkpoint that can stop the eukaryotic cell cycle in case of ongoing DNA replication problem or DNA damage.

Discussion
Primitive cell cycle regulatory processes might have evolved somewhere between the first and the last common ancestor of eukaryotes to ensure the once-and-only once replication of multiple chromosomes and the accurate partitioning of sister chromatids to the incipient daughter cells. Of equal importance, chromosome replication had to be coordinated with overall cell growth, chromosome replication had to be blocked if the genome were damaged in any way, and mitosis had to be delayed until the cell had acquired sufficient energy stores to complete the process. In particular, cells evolved a system that stops progression through the cell cycle when something goes wrong with the genetic material. But when high levels of ATP push all kinases to be active, and thereby induce cell division, it is challenging to stop cell-cycle progression even if DNA is damaged. The observed flip in the roles of Wee1 kinase and Cdc25 phosphatase could have been sufficient to deal with this challenge. The kinase Wee1 lined up with interphase factors, while the phosphatase Cdc25 teamed up with mitotic factors. In this way, cell division can be arrested in case of any damage to the DNA or if DNA replication is still ongoing. The ability to decouple DNA replication and division became necessary only with the emergence of multiple chromosomes in eukaryotes, explaining why all members of the Cdk1-Wee1-Cdc25 network appeared with the emergence of eukaryotes. Wee1 and Cdc25 are unique enzymes. They are not functionally related to other enzymes in the network, but they are cornerstones of the regulatory system. There is certainly a lot left to be investigated about the evolution of the cell cycle regulatory network in eukaryotes 89,97 . We showed here how a twist in the Cdk1-Wee1-Cdc25 system allowed the cell cycle to be effectively arrested in an energy rich environment.
The observed twist from the general rule of kinase-phosphatase activity to meet the demands of a functional checkpoint might not be unique to the regulation of the G2/M transition. Certainly the G1/S transition is similarly controlled by an isoform of Cdc25 phosphatase 98 and the mitotic checkpoint was also proposed to be influenced by complex interactions of kinases and phosphatases 99 . The above-mentioned regulation of AMPK by energy level also shows that the AMP-activated protein kinase is inhibited by ATP and activated by a phosphatase, which is again a twisted system. There might be several other examples, where such a flip in kinase/phosphatase regulation ensures that a threshold can be set, before a crucial cellular transition could occur. It is tempting to speculate that the observed twist in kinase-phosphatase interaction could be a more general mechanism for establishing checkpoints and adjustable thresholds in cellular signalling.

Data availability
The data underlying this article are available in the article and in its online supplementary material. www.nature.com/scientificreports/