A Dynamical Framework for the All-or-None G1/S Transition

Establishing a System to Analyze the Dynamics of the G1/S Transition in Human Cancer Cells

We first sought to follow the dynamics of Cyclin and p27^Kip1 expression in HeLa cells at the G1/S transition. We genetically modified bacterial artificial chromosomes (BACs; Ciotta et al., 2011, Hutchins et al., 2010) to generate GFP-tagged p27^Kip1, CyclinE1, and CyclinA2, under the control of their endogenous promoters. These constructs were used to generate clonal cell lines for this study (Figure S1A). Expression of all three reporters mimicked the localization and expression of their endogenous counterparts (Figures 1 and S1B–S1D), indicating that the GFP tag did not affect their cell-cycle-mediated regulation. Moreover, the GFP tag did not prevent S-phase entry or cell-cycle progression (Figures S1E and S1F) and did not affect protein degradation rates (Figures S1G and S1H; timely degradation of CyclinA2-GFP has been validated in den Elzen and Pines, 2001). Thus, our system accurately reflects Cyclin and p27^Kip1 expression throughout the G1/S transition.

To correlate Cyclin and CKI expression with the G1/S transition, we co-imaged each GFP tagged protein with LSS2-mKate-PCNA (Figures 1, S1C, and S1D; Movies S1, S2, and S3). Fluorescently labeled PCNA is a robust and well-used marker of DNA replication in mammalian cells (Essers et al., 2005, Held et al., 2010, Leonhardt et al., 2000, Pomerening et al., 2008) and does not interfere with cell-cycle progression in our system. When cells enter S phase, PCNA fluorescence rapidly increases and then switches to a punctate localization (Figures 1A–1D, S1C, and S1D). Since the duration of G1 varies between cells, LSS2-mKate-PCNA localization can act as a marker to pseudo-synchronize cells at the G1/S transition, which we define as time zero (Figures 1A–1C).

We next sought to define the temporal regulation of each of the GFP tagged G1/S regulators throughout the G1/S transition. Consistent with previous observations made in HeLa cell populations by western blotting (Carrano et al., 1999), we found that p27^Kip1-GFP is absent after cell division, accumulates rapidly, and is then degraded, starting approximately 5–10 hr after mitotic exit (Figure S1D). Through live single cell imaging, we observed a decelerating pattern of p27^Kip1-GFP accumulation during G1 that strongly suggests that p27^Kip1 degradation is turned off after cell division. By imaging p27^Kip1-GFP and LSS2-mKate PCNA in the same cell, we saw that p27^Kip1-GFP was very rapidly degraded immediately prior to S-phase entry (Figures 1A and S1D). Western blotting of synchronized cell lysates did not show this rapid degradation because in a cell population cells enter S phase at different rates (Figure S1G), and thus, p27^Kip1 degradation appears slow.

We observed that CyclinE1-GFP accumulates prior to the G1/S transition and is only degraded after S-phase entry (Figure 1B), consistent with previous reports of CyclinE levels in HeLa cell populations determined by western blotting of lysates from synchronized cells (Dulić et al., 1992, Koff et al., 1992). Single cell imaging also revealed that CyclinE1-GFP levels reach a short plateau immediately after S-phase entry, consistent with the plateau reached by endogenous CyclinE as detected by western blotting (Yuan et al., 2014). Moreover, accumulation of CyclinE1-GFP occurred in an accelerated fashion.

In contrast to CyclinE1-GFP, CyclinA2-GFP remained low during G1, despite the fact that their common transcription factor (E2F) was active during G1 (Figure 1C). Imaging of CyclinA2-GFP showed that CyclinA2 protein only began to accumulate in a linear manner after the G1/S transition (Figure 1C), consistent with previous reports from western blotting of HeLa cell lysates (Pines and Hunter, 1990). This suggests that CyclinA2 is degraded during G1.

Notably, the levels of CyclinE and CyclinA alone do not reveal Cdk2 activity since Cyclin binding alone is not necessarily sufficient to activate Cdks (reviewed in Morgan, 1997). To measure Cdk2 activity, we generated a Cdk2 activity sensor (CDK2L-GFP), a slightly longer version of the recently published DHB-Ven Cdk2 sensor (see Experimental Procedures; Gu et al., 2004, Spencer et al., 2013). Briefly, this is a fluorescent sensor of Cdk2 activity based on the C-terminal fragment of DNA Helicase B. When phosphorylated by Cdk2, the sensor leaves the nucleus. By measuring the cytoplasmic:nuclear ratio of CDK2L-GFP, we can obtain an estimate of Cdk2 kinase activity from individual cells. First, we validated that CDK2L-GFP was a specific reporter of Cdk2 activity in HeLa cells (Figures S2A–S2I). As judged by our CDK2L-GFP sensor, Cdk2 kinase activity abruptly increased at S-phase entry, followed by a second phase of activity increase that occurs at a relatively slower rate (Figure 1D; Movie S4). We saw similar behavior if we used the DHB-Ven Cdk2 activity sensor (Spencer et al., 2013) in HeLa cells (Figure S2J). The abrupt increase in kinase activity was precisely correlated with the rapid decrease in p27^Kip1-GFP (Figures 1A and 1D), suggesting that degradation of p27^Kip1 may be rate limiting in Cdk2 activation and S-phase entry. Immediately after S-phase entry, we consistently observed a transient decrease in Cdk2 activity (Figures 1D and S2J). In CyclinE1/2-depleted cells, this decrease is more pronounced (Figure 3D). However, what causes this dip in Cdk2 activity is currently not known and warrants further investigation. Taken together, these observations describe key G1/S signaling events in single cells throughout the G1/S transition.

Developing a Dynamical Model of the G1/S Transition

To understand how the observed dynamics of Cyclin, CKI and Cdk2 activity emerge from the underlying molecular network, we next generated a dynamical model of the G1/S transition in human cells based on our live imaging data. Until now, the dynamical data necessary to parametrize such a model have not been available, as the current literature provides information only on the molecular mechanisms controlling the G1/S transition (Kohn, 1999). This G1/S molecular network involves multiple interactions between Cdk2:CyclinA, Cdk2:CyclinE, p27^Kip1, and their regulators (Figure 2A).

We have used prior knowledge of the G1/S molecular network to construct a stochastic mathematical model of the G1/S transition (see Supplemental Information for a detailed description of rate expressions). Our model contains a considerable number of kinetic parameters (25) that are largely unknown in the current literature, except for the half-lives of some mRNAs and proteins (Schwanhäusser et al., 2011). The unknown rate constants were estimated by comparing deterministic simulations of the model (Figure 2C) with averaged time courses of pseudo-synchronized cell-cycle regulators (Figures 2B and S3A) in a brute-force approach.

In our model, active E2F leads to an accumulation of CyclinE protein early in G1 (Figures 2C and S3B), in the absence of functional pRb in HeLa cells (Moody and Laimins, 2010, Schwarz et al., 1985). However, CyclinE accumulation is insufficient to drive progression into S phase as Cdk2:CyclinE activity is inhibited by its stoichiometric CKI, p27^Kip1, whose degradation, dependent on Cdk2 and SCF^Skp2 activities (Sheaff et al., 1997), is slow at low Cdk2 activity and low Skp2 levels. Once sufficient CyclinE accumulates, Cdk2:CyclinE activity overcomes p27^Kip1 inhibition. The activated Cdk2:CyclinE targets p27^Kip1 for proteasome-dependent degradation through phosphorylation of T187 on p27^Kip1 (Müller et al., 1997, Sheaff et al., 1997, Vlach et al., 1997). With the degradation of p27^Kip1, more Cdk2:CyclinE is released, and p27^Kip1 degradation is further enhanced. The fast positive feedback results in the abrupt degradation of p27^Kip1. After the abrupt degradation of p27^Kip1, Cdk2:CyclinE kinase becomes activated (Figures 2C and S3B).

In our model, despite the activation of CyclinA protein synthesis, the level of CyclinA is kept low in G1 phase by APC/C^Cdh1-mediated degradation (Figures 2C and S3B; Geley et al., 2001, Lukas et al., 1999, Sørensen et al., 2001). Accumulation of CyclinA does not start until inactivation of APC/C^Cdh1. At the end of G1, APC/C^Cdh1 is inactivated due to the accumulation of Emi1 (Hsu et al., 2002, Miller et al., 2006) and phosphorylation of Cdh1 by Cdk2 (Keck et al., 2007, Lukas et al., 1999). This leads to accumulation of Skp2 (Bashir et al., 2004, Wei et al., 2004). The degradation of phosphorylated p27^Kip1, which is dependent on SCF^Skp2-dependent poly-ubiquitylation, is further accelerated by Skp2 accumulation in late G1 (Figure S3C).

All of these events—p27^Kip1 degradation, Cdk2 activation, and CyclinA accumulation—coincide with the G1/S transition. This leads to the phosphorylation-dependent degradation of CyclinE protein by Cdk2 (Won and Reed, 1996). A linear combination of the model-calculated free Cdk2:Cyclin complexes (Figure 2C) therefore approximates the Cdk2 activity, as measured by our Cdk2 sensor (Figure 2B).

Our model explains the switch-like characteristic of the G1/S transition by mutual inhibition among cell-cycle regulators, with a central role for p27^Kip1. The abrupt degradation of p27^Kip1 has a critical threshold and fast protein turnover requirement. In our model, the threshold originates from the antagonistic interaction between p27^Kip1 and Cdk2:Cyclin complexes, which creates a bistable switch with the following characteristics. (1) The steady-state level of p27^Kip1 can be either high or low depending on CyclinE levels, but intermediate steady-state levels are unstable (Figure 2D). (2) The transition from a high to low p27^Kip1 level takes place at a higher threshold of CyclinE than the opposite transition. Therefore, G1/S control possesses the property of hysteresis. (3) The threshold of CyclinE at which transitions occur is dependent on the levels of Emi1 that inhibit APC/C^Cdh1 and thereby upregulate Skp2 (Figure 2D). At the beginning of the cycle when both CyclinE and Emi1 levels are low, the thresholds are high, p27^Kip1 is attracted to the upper steady state, and its level rises (red curve in Figure 2D). (4) During G1, E2F-dependent transcription drives the increase in both CyclinE and Emi1 protein levels. Thus, CyclinE and Emi1 could act synergistically to remove p27^Kip1. CyclinE levels reach the upper threshold at a relatively low level of Emi1 that drives the cell through G1/S into a state of low p27^Kip1 (Figure S3D). (5) After the G1/S transition, CyclinE levels start to decrease, but the cell stays in the low p27^Kip1 regime because the thresholds are also decreasing through Emi1 accumulation (Figure S3D). The continuous decrease of the thresholds makes G1/S an irreversible transition in the model.

Critically, our stochastic model captures the observed dynamics of G1/S regulators in HeLa cells (compare Figure S3A with S3B) and thus provides a suitable framework by which to make novel predictions regarding the mechanisms that underlie the dynamical nature of the G1/S transition.

Degradation of p27^Kip1 Is Driven by Cdk2 Activity

Our model of the G1/S regulatory network attributes the abrupt decrease of p27^Kip1 to fast protein turnover regulated by direct and indirect double-negative feedback loops between p27^Kip1 and Cdk2. Specifically, activated Cdk2 promotes p27^Kip1 degradation. Our model also implies that Cdk2 is required to keep the level of p27^Kip1 low after G1/S. Inhibition of Cdk2 should lead to re-accumulation of p27^Kip1 (Figure S3E). To test this prediction, we treated HeLa cells, arrested in early S phase by a thymidine block, with a Cdk1/2 inhibitor. In S-phase-arrested cells, p27^Kip1 levels are negligible. However, upon addition of Cdk1/2 inhibitor, p27^Kip1 rapidly re-accumulated (Figure S3F). This rapid re-accumulation of p27^Kip1 suggests that its synthesis (transcription and translation) is still active during DNA replication. Moreover, the low level of p27^Kip1 after G1/S transition is indeed maintained by high Cdk2 activity promoting p27^Kip1 degradation. Therefore, our model correctly predicts that a low level of p27^Kip1 is maintained throughout S phase by high Cdk activity.

CyclinE Controls the Timing of the G1/S Transition

Stochastic model simulation with reduced levels of residual CyclinE synthesis suggests that the G1/S transition is delayed if CyclinE levels are depleted in G1 (compare the red and gray p27^Kip1 curves on Figure 3A). However, the rapid degradation of p27^Kip1 and the abrupt activation of Cdk2 are conserved even after CyclinE depletion. The abruptness of p27^Kip1 degradation in the absence of CyclinE can be attributed to the mutual antagonism between Cdk2:CyclinA and its inhibitor p27^Kip1 (Figure 2A). The steady-state level of p27^Kip1 shows a similar Z-shaped dependence on the levels of CyclinA (not shown). The two CyclinA thresholds, where p27^Kip1 decreases or rises, are Emi1 dependent because Emi1 inhibits APC/C^Cdh1 to upregulate Skp2 levels (Figure 3B). The model predicts that the “path” during the G1/S transition is different for control siRNA-treated cells (Figure S3D) and CyclinE-depleted cells (Figure 3B). In control siRNA cells, CyclinE and Emi1 accumulate simultaneously. Since Emi1 is low, APC/C^Cdh1 is largely active, and Skp2 level is low. Therefore, a large amount of CyclinE is needed (>0.6 arbitrary units [a.u.]) to trigger p27^Kip1 degradation. In CyclinE-depleted cells, Emi1 accumulates first, leading to APC/C^Cdh1 inactivation and subsequent accumulation of CyclinA. In the presence of a high level of Emi1, inactive APC/C^Cdh1 and a high level of Skp2, a low level of CyclinA (approximately 0.2 a.u.) is sufficient to trigger p27^Kip1 degradation.

Consistent with our model prediction, siRNA-mediated depletion of CyclinE delayed the degradation of p27^Kip1, activation of Cdk2, and initiation of S phase (compare the gray with the red and black curves on Figures 3C and 3D, respectively; see also Figures S4A and S4B). In addition, the abrupt degradation of p27^Kip1 and abrupt activation of Cdk2 remained intact after CyclinE depletion (Figures 3C and 3D). Together, our results suggest that although CyclinE is dispensable for the initiation of DNA replication (Geng et al., 2003) CyclinE does play a role in the timing the G1/S transition and thus is rate limiting.

CyclinA Is Dispensable for the Timing of G1/S Transition

Stochastic model simulation with reduced levels of CyclinA suggests no change in the timing of the G1/S transition (compare the red and gray p27 curves on Figure 4A). In addition, the simulation suggested that neither p27^Kip1 degradation nor Cdk2 activation showed any delay relative to control cells and that these events also kept their close association with the G1/S transition. These simulation results are consistent with our experimental testing, in which CyclinA is depleted by siRNA (compare the gray with the red and black curves on Figures 4B and 4C, respectively; see also Figures S4A and S4C).

Since CyclinA does not control the timing of the G1/S transition, the observation that different cells enter S phase at different times is unlikely due to their different levels of CyclinA. Based on our stochastic simulations (Figure 4A), we predict that the stochastic expression of other G1/S regulators (for example CyclinE or Emi1) is responsible for the variability in timing of the G1/S transition in CyclinA-deficient cells.

Emi1 Ensures an Irreversible G1/S Transition

One way to ensure that DNA is replicated once and only once per cell cycle is to make the G1/S transition difficult to reverse. Intuitively, we might predict that by binding and activating Cdk2 CyclinA could maintain an S-phase state after CyclinE is degraded. However, our model simulations predict that CyclinA depletion in S-phase cells is not sufficient to trigger re-accumulation of p27^Kip1 and thus reset cells into G1. In contrast, our model suggests that Emi1 is both necessary and sufficient to maintain the post-G1 state. Specifically, our model predicts that increasing levels of Emi1 during S phase decrease the amount of CyclinE required to maintain p27^Kip1 degradation (Figures 2D and S3D). If we remove Emi1 from S-phase cells in the model, APC/C^Cdh1 is activated, CyclinA and Skp2 are degraded, and p27^Kip1 and CyclinE re-accumulate. Later, increasing CyclinE causes a second round of p27^Kip1 degradation, and the cells could re-enter S phase once again (Figures 5A and S5A).

To test our model prediction, we synchronized cells in early S phase (with a double thymidine block) and depleted Emi1 by RNA interference during release from the first thymidine block, such that Emi1 would be depleted in the next cell cycle (Figure S5B; see Supplemental Experimental Procedures). Control siRNA-treated cells undergo mitotic cycling, confirmed by fluorescence-activated cell sorting (FACS) (Figure 5B) and the temporal patterns of cell-cycle regulators (see Figures S5C–S5F). In contrast, Emi1 depletion leads to re-replication of DNA, as previously reported (Figure 5B; Di Fiore and Pines, 2007, Machida and Dutta, 2007). This suggests that Emi1 is required to maintain an irreversible G1/S transition.

Consistent with our model prediction, the experimental data clearly show a short-lived increase then degradation of CyclinA2-GFP (Figure 5C), followed by transient accumulation of p27^Kip1-GFP (Figure 5D), and re-accumulation of CyclinE1-GFP (Figure 5E). This accumulation of CyclinE1 protein is likely due to a combination of a low level of Cdk2 activity since CyclinE1-GFP is stabilized when Cdk2 activity is inhibited (Figure S6A) and high E2F activity as cells return to a G1-like state. Direct measurements of Cdk2 activity in Emi1-depleted cells show an initial small increase in kinase activity after release from the thymidine block, followed by a decrease. In the majority of cells, this decrease is small and is followed by a plateau in Cdk2 levels at an intermediate level of between 1 and 1.5 (Figure 5F). However, in some cells, this dip in Cdk2 activity is more pronounced. The activity of Cdk2 in any individual Emi1-depleted cell will depend on (1) the level of Emi1 depletion and (2) the relative amounts of CyclinE and CyclinA when Emi1 is depleted in that cell. Therefore, cells with a larger dip in Cdk2 activity may reflect cells with a better Emi1 depletion and/or a higher ratio of CyclinA to CyclinE when Emi1 is depleted.

This begs the question of how DNA re-replicates after Emi1 depletion. Notably, we do not observe endoreduplication after Emi1 depletion (Figure 5B; no discrete peak is seen at 8n; at 200K on the x axis). If we image PCNA dynamics in cells depleted of Emi1, we can observe nuclei switching between G1 and S in the absence of mitosis and nuclei increasing in size (Figure S6B; Movies S5 and S6). However, PCNA foci in Emi1-depleted cells are not discrete, as in control siRNA-treated cells, and S-phase entry is more noticeable by changes in the “texture” of PCNA localization throughout the nucleus (Movie S6). This implies that after Emi1 depletion, cells do not progress through “early,” “mid,” and “late” DNA replication stages, but instead likely sit at the boundary of G1 and S, refiring some DNA replication origins. Loss of Geminin after Emi1 depletion (Di Fiore and Pines, 2007, Machida and Dutta, 2007) would permit origin refiring. This is also consistent with recent data suggesting uncontrolled reactivation of replication origins after Emi1 depletion (Neelsen et al., 2013). Altogether, our data suggest that S-phase cells depleted of Emi1 can transiently return to a G1 state (low CyclinA, high p27^Kip1, accumulating CyclinE) before recommitting to DNA synthesis, supporting our model prediction that Emi1 maintains the irreversibility of the G1/S transition.

Generation of Fluorescently Tagged BACs and Reporters

BACs used in this study were ordered from BACPAC Resource center (CHORI). BACs used are p27^Kip1 RP11-70N14, CCNE1 RP11-345J21, and CCNA2 RP11-768B18. All BACs were verified by PCR using verification primers as listed on the Mitocheck website (http://www.mitocheck.org). Proteins of interest were tagged using BAC Recombineering at their C-termini with localization and affinity purification (LAP) tags that contain GFP and an affinity purification tag (Poser et al., 2008). Primers used to tag BACs were those listed on the Mitocheck BACFinder website (http://www.mitocheck.org/cgi-bin/BACfinder). Successful recombination and tagging of genes was confirmed by PCR followed by DNA sequencing.

PCNA was tagged at the N terminus with the modified LSS2-mKate fluorophore to generate LSS2-mKate-PCNA (Piatkevich et al., 2010).

To measure CDK2 activity, we used a similar strategy, now described in Spencer et al. (2013). We cloned the C-terminal PSLD region of Human DNA Helicase B (DHB; 957-1087 amino acids of DHB) into pIRES-GFP Puro3 to generate CDK2L-GFP. Note that this sensor is slightly longer than that used in Spencer et al. and is similar to the one first described in Gu et al. (2004).

Live Cell Imaging

Live cell imaging was performed using a Zeiss LSM710 confocal, with a 40× oil objective, NA 1.3, or a High-content Opera Spinning disk confocal microscope (PerkinElmer), with a 40× water objective, NA 0.9. All live cell imaging was carried out in a humidified environmental chamber maintained at 37°C and 5% CO₂.

Quantification of Reporters

All imaging data was quantified in Volocity (PerkinElmer). To find dynamic changes in GFP intensities, cell nuclei were tracked using the LSS2-mKate-PCNA signal and the Shortest Path algorithm. A maximum distance of 9 μm between nuclei in consecutive timeframes was used to eliminate erroneous tracks. For p27^Kip1-GFP, CyclinE1-GFP, and CyclinA2-GFP, nuclear levels of protein were quantified since all three proteins had an exclusively nuclear localization during G1 and S. In each cell, the average value of pixel intensities was calculated, and average background fluorescence was subtracted. All fluorescence intensities were normalized to peak levels. Timing of S-phase entry was determined manually from PCNA fluorescence images. Specifically, S-phase entry is defined by a sharp increase in PCNA intensity, often coinciding with the appearance of nucleoli.

For CDK2L-GFP and DHB-Ven quantification, nuclear intensity was measured by taking a region of interest (ROI) inside the nucleus, and cytoplasmic intensity was measured as the ring region around the nucleus. Cdk2 activity was then measured as the ratio between the ring GFP intensity to the nuclear GFP intensity.

Additional experimental procedures are provided in Supplemental Information.