Slower CDK4 and faster CDK2 activation in the cell cycle

Summary

Dysregulation of cyclin-dependent kinases (CDKs) impacts cell proliferation, driving cancer. Here, we ask why the cyclin-D/CDK4 complex governs cell cycle progression through the longer G1 phase, whereas cyclin-E/CDK2 regulates the short G1/S phase transition. We consider available experimental cellular and structural data including cyclin-E high-level burst, sustained duration of elevated cyclin-D expression, and explicit solvent molecular dynamics simulations of the inactive monomeric and complexed states to establish the conformational tendencies along the landscape of the distinct activation scenarios of cyclin-D/CDK4 and cyclin-E/CDK2 in the G1 phase and G1/S transition of the cell cycle, respectively. These lead us to propose slower activation of cyclin-D/CDK4 and rapid activation of cyclin-E/CDK2. We provide the mechanisms through which this occurs, offering innovative CDK4 drug design considerations. Our insightful mechanistic work addresses a compelling cell cycle regulation question and illuminates the distinct activation speeds in the G1 versus G1/S phases, which are crucial for function.

Graphical Abstract

eTOC Blurb:

Zhang et al. discover different activation speeds of cyclin-D/CDK4 and cyclin-E/CDK2, crucial for the timescales of the cell cycle progression stages. These findings suggest harnessing the differential speeds’ mechanisms in cancer drug development, collectively providing a deeper understanding of cell cycle regulation and opening new pathways for targeted drug discovery.

INTRODUCTION

Cyclin-dependent kinases (CDKs) are a family of serine/threonine kinases essential for cell cycle regulation.^1–3 Among them, CDK2 and CDK4 are crucial for the initiation of DNA replication and progression from the G1 to the S phase. The kinases are activated by binding to their respective regulatory cyclin partners; cyclin-A and cyclin-E to CDK2 and cyclin-D to CDK4. Cyclins are synthesized and degraded in a cyclical fashion throughout the cell cycle and cyclin/CDK complexes are further regulated by phosphorylation and dephosphorylation events.^1,4 Aberrant expressions of cyclin-D, cyclin-E, and cyclin-A are common in cancer cells.⁵ Cyclin/CDK complexes undergo specific changes in their activity during different stages of the cell cycle, ensuring proper cell cycle progression. Dysregulation of CDK activity is frequently observed in cancer cells and plays critical roles in the development and progression of cancer and neurodegenerative disorders.^6–11 Increased CDK4 activity in cancer cells can be due to genetic mutations, amplification, and overexpression of upstream nodes in the signaling pathways and in its regulatory cyclin-D. CDK2 and CDK4 can be constitutively active and overexpressed including in adrenocortical carcinomas, breast, lung, and prostate cancers,^6,12–16 making them promising drug discovery targets.^12–16 Inhibiting them can impede cell cycle progression, curtailing cell division and proliferation.¹⁷ CDK4 plays a crucial role in the G1 phase–the longest phase of the cell cycle.^9,18–21 CDK2, activated by cyclin-E binding in the G1/S transition also acts in DNA damage repair.²² Significantly, although mutations in CDK2 and CDK4 genes have been identified in some cancer cells, different than other kinases such as PI3K²³, their frequency is low, especially in the catalytic kinase domain, as shown in The Cancer Genome Atlas (TCGA) and Genomics Evidence Neoplasia Information Exchange (GENIE) databases,⁷ emphasizing their severe outcome.

Numerous studies have shed light on the molecular mechanisms underlying the regulatory roles of CDKs in cell cycle progression and their inhibition by small molecules.^{15,17,22,24–28} The CDK interacting protein/Kinase inhibitory protein (CIP/KIP) inhibitors bind to CDK2 complexes, hampering the kinase activity, while inhibitors of CDK4 (INK4) bind to CDK4 or CDK6, preventing their interaction with cyclin-D.^17,26,29 CDK4 and CDK6 inhibitors (abemaciclib, palbociclib, and ribociclib) have shown efficacy in treating estrogen receptor-positive breast cancer and are now approved for clinical use.^25,30–32

In the canonical cell cycle activation pathway, CDK4 is activated in the G1 phase, while CDK2 is involved in the G1/S transition and throughout the S-phase.³³ Active CDK4 in complex with cyclin-D partially phosphorylates the retinoblastoma protein (Rb), disrupting the Rb/E2F interaction and releasing E2F transcription factors, which then trigger the expression of cyclin-E. Subsequent CDK2 activation by cyclin-E leads tohyperphosphorylation of the Rb protein.^34,35

While the sequences of CDK2 and CDK4 differ (Fig. 1a), they share the typical protein kinase conformation (Fig. 1b). The crystal structure of active CDK2 in complex with cyclin-E is available, while the active CDK4 structure in complex with cyclin-D has long been elusive. CDK4 retains a stable inactive conformation even when cyclin-bound and phosphorylated on the activation segment, suggesting that binding of the protein substrate and other factors are required for full activation (Fig. 1c).³⁶ Gharbi et al.³⁷ have recently determined the crystal structure of active CDK4 in complex with cyclin-D. Recent studies using molecular dynamics (MD) simulations and other computational approaches reported differences in the conformational landscapes of CDK2 and CDK4 complexes as well as the effects of CDK inhibitors on the activation of these complexes.^38–41 These studies suggested that CDK2 and CDK4 have distinct conformational dynamics and allosteric interactions with their regulatory cyclins and peptide inhibitors, which may contribute to their different activation kinetics.

Figure 1. Sequence and structural alignments of CDK2 and CDK4.

(a) Sequence and structural alignment of CDK2 and CDK4, showcasing their similarities and differences. (b) Superimposition of the crystal structures of CDK2 and CDK4. Key domains and residues are labeled. (c) In silico modeling of the inactive CDK2 and CDK4 in complex with their cognate cyclin partners, cyclin-E, and cyclin-D, respectively. The activation loops, crucial for kinase activity, are highlighted in pink.

Despite their vital importance, and the significant progress, the precise mechanisms by which CDK4 and CDK2 are activated by cyclin-D in the G1 phase and cyclin-E in the G1/S transition of the cell cycle, respectively, remain unresolved. The duration of the G1 and the G1/S transition in the cell cycle differ. G1 is long. G1/S is short. The expression levels and concentration of the respective cyclins differ as well. Cyclin-E expression level is regulated by a positive feedback loop, peaking, and falling over a short time span, which is not the case for cyclin-D. CDK4 has been traditionally deemed as the more accessible for drug discovery. Thus, in this work, we aim to understand why cyclin-E/CDK2 have been recruited for the transition and cyclin-D/CDK4 for the G1 phase. We query their conformational properties and landscapes and ask which states and activation pathways are favored. We focus on the inactive conformations of CDK2 and CDK4 in their monomeric states, as well as in complex with their respective cyclin partners (Figs. 1 and S1) to elucidate the effects of ATP and cyclin binding on the conformational shift from the inactive toward the active states of these kinases. Note that even in the case of the simulations of CDK2 and CDK4 in complex with their cyclin partners, we still use the inactive conformations of CDK2 and CDK4 (with the OUT αC-helix and collapsed loop conformation) to observe their tendency to shift to the active state. Explicit solvent MD simulations of these systems were performed for inactive CDKs with and without ATP in the active site. Our idea has been to model the activation of cyclin-E/CDK2 and cyclin-D/CDK4 each step of way along the conformational energy landscape to determine the preferred sequence of the activation events. Kinase activation often occurs on the time scale of microseconds or longer. To compensate for this inherent limitation of MD simulations, we simulate CDK conformational states at multiple stages along the activation pathway and observe their tendency to undergo a conformational change. We can then link these different trajectories and infer their preferred activation steps. We exploit this strategy to decipher the mechanistic path and test our thesis. It leads us to determine what we perceive as the fundamental guiding principle shaping the activation pathways of the two cell cycle systems: their rates of activation. Activation of cyclin-E/CDK2 which executes a cell cycle stage transition is fast, whereas that of cyclin-D/CDK4 is slow.

Here we analyze the conformational landscape of CDK2 and CDK4 upon ATP and cyclin binding and propose activation steps for these two CDK complexes. Our observations support CDK2 preferentially binding ATP first then cyclin-E, activating rapidly, while CDK4 favors binding cyclin-D first then ATP, resulting in a slower activation process. CDK2 has a more flexible activation loop (A-loop) and greater tendency to facilitate ATP loading compared to CDK4, predisposed to shift toward its active conformation more readily. We attribute their differences to the longer CDK4 loops connecting the β3-strand to the αC-helix, and the β4-strand to the β5-strand, which block the αC-helix from moving inward. A key driving force of CDK4 slower activation is the stronger hydrophobic interactions between the A-loop and the αC-helix. This is bolstered by the interaction of the longer CDK4 loops which, upon cyclin binding, can thwart the formation of the catalysis-ready state compared to CDK2. Our studies pioneer the connection between the cyclins-CDKs mechanisms and their roles in cell cycle function, contribute to a deeper understanding of the activation mechanisms and their linkage to cell cycle phases and timescales, and potentially informing the development of novel therapeutics.

RESULTS

CDK2 is more pre-organized compared to CDK4 to facilitate ATP loading

Protein kinases play crucial roles in regulating cellular processes by transferring phosphate groups from ATP molecules to specific protein substrates. The ATP-loading ability of the kinases is one of the determinants for their efficiency in catalysis. Kinases allowing easier ATP entry and more stable binding may be more likely to efficiently catalyze reactions. The A-loop can block or allow ATP entry and plays an important role in regulating the transfer of phosphate groups. In our simulations, the monomeric apo states of CDK2 and CDK4 (hereafter CDK2^Apo and CDK4^Apo, respectively) adopted an inactive conformation with the collapsed A-loops, which obstructs access to both the ATP-binding pocket and substrate binding site.⁴² For CDKs to effectively carry out their phosphorylation functions, the A-loops need to be repositioned, and the ATP-binding pocket must expand to accommodate ATP loading. Recent studies in Src kinases implied the flexibility of the A-loop and the dimensions of the ATP-binding pocket are crucial factors influencing the ability of kinases to load ATP.^43,44 Building upon these findings, we assess the flexibility of the A-loops in both CDK2^Apo and CDK4^Apo. Despite their structural similarities, they have differences in their A-loop sequences and dynamics (Fig. 2a). Superimposed A-loop conformations over the simulation trajectories illustrate that the monomeric CDK2^Apo has more flexible A-loop than the monomeric CDK4^Apo. The rootmean-squared-fluctuations (RMSFs) verify that CDK2^Apo shows increased fluctuation of the A-loop as compared to CDK4^Apo (Fig. 2b). The different A-loop dynamics may be caused by its different length, which is longer in CDK2 than in CDK4. A flexible A-loop in the apo state implies a greater tendency for the A-loop to adopt an extended conformation. The A-loop flexibility can significantly change once the kinase is active, where a more rigid A-loop may be linked to enhanced catalytic activity. Based on our observations, apo CDK2 is more likely to extend its A-loop, allowing ATP entrance, which can explain recent NMR observations for the sampled CDK2^Apo conformations with the A-loop OUT state.⁴⁵ Our results show that CDK2^Apo is more pre-organized to facilitate ATP loading, suggesting a greater readiness for activation and substrate phosphorylation compared to CDK4.

Figure 2. Activation loop of CDK2^Apo has a greater flexibility than CDK4^Apo, facilitating ATP entrance.

(a) Superimposition of snapshots of the monomeric inactive CDK2^Apo reveals that the A-loop of CDK2 exhibits greater flexibility as compared to CDK4. This increased flexibility may contribute to differences in ATP binding and activation kinetics between these two kinases. (b) Root-mean-squared-fluctuation (RMSF) and sequence alignment of the A-loops for CDK2 and CDK4. Yellow highlighted regions indicate the phosphorylation sites, which play a crucial role in modulating kinase activity. The RMSF values demonstrate the extent of flexibility within the A-loops, further illustrating the inherent structural differences between CDK2 and CDK4.

Conformational flexibility of CDK2 decreases on ATP- and cyclin-binding in contrast to CDK4

To evaluate the distinct dynamics of CDK2 and CDK4 upon ATP- and cyclin-binding, we calculated the potential of mean force (PMF) with two reaction coordinates. The PMF helps us to understand the stability and conformational state of these kinases. The PMF of proteins with lower ΔG_PMF indicates that the protein conformation is very stable and vice versa. The first reaction coordinate reflects the “height” of the ATP-binding cleft. The residue-based distances between the valine (V18 in CDK2 and V20 in CDK4) in the roof and the leucine (L134 in CDK2 and L147 in CDK4) at the bottom of the active-site cleft were calculated as proxies for the ATP-binding site’s open/closed conformations. The Val-Leu distance is calculated based on the corresponding Cα atoms. The second reaction coordinate relates to the propensity of the kinase to transition from an inactive to active conformation, which is centered on the distance between Lys (K33 in CDK2 and K35 in CDK4) on the β3-strand and Glu (E51 in CDK2 and E56 in CDK4) on the αC-helix. This reaction coordinate informs about the tendency to form a critical salt bridge associated with a kinase activity. For the Lys-Glu salt bridge pair, the distance between the nitrogen atom on the side chain of Lys and one of the oxygen atoms on the side chain of Glu was calculated. As a typical protein kinase, CDKs adopt an IN αC-helix in the active state and an OUT αC-helix in the inactive state.¹ The transition from the inactive to the active conformation involves ~30° rotation of the αC-helix.⁴⁶ This rotation results in forming a conserved salt bridge between a glutamic acid in the αC-helix and a lysine in the β3-strand. The salt bridge stabilizes the IN αC-helix in the active conformation. In the catalytic active state, the salt bridge stabilizes the γ-phosphate of ATP for phosphate transfer, while in the inactive state, the salt bridge is disrupted, inhibiting the catalytic reaction.^27,47 We aim to monitor the tendency of the αC-helix positioning either toward the OUT or IN position, rather than observing the OUT to IN transition or vice versa, which is beyond the computationally accessible time scale. We observed that the αC-helix of the inactive CDK2 has a greater tendency to move inward upon ATP loading and cyclin binding (Fig. S4). The superimposed snapshots show that the conformational fluctuations of CDK2^Apo are greater than that of the cyclin-E/CDK2^ATP complex (Fig. 3a), while the opposite is true for CDK4, where the conformational fluctuations of cyclin-D/CDK4^ATP are greater than those of CDK4^Apo (Fig. 3b). Our results also reveal distinct PMF landscapes for CDK2 and CDK4. Specifically, the PMF of CDK2 is more localized upon ATP- and cyclin-binding, as indicated by the decreased ΔG_PMF value when comparing the PMF of CDK2^Apo and cyclin-E/CDK2^ATP complex, suggesting a strong propensity for CDK2 to shift toward its active conformation. Conversely, cyclin-D/CDK4^ATP shows a “shallowing” (more uniform) of the PMF upon ATP- and cyclin-binding, as indicated by the higher ΔG_PMF value when compared to that of CDK4^Apo, indicating a broader range of conformational states on its way to activation and a weaker propensity to shift toward its active state. It shows that the inactive CDK2 is more pre-organized to facilitate the transition to the active state upon ATP loading and cyclin-binding than CDK4. Our analysis of these two reaction coordinates further shows that CDK2^Apo has an open ATP-binding site (Fig. S4a, blue curve), which facilitates ATP entry into the binding site. Once ATP enters the active site cleft, the pocket is closed (CDK2^ATP). In contrast, the monomeric CDK4^Apo has a relatively closed ATP-binding site (Fig. S4b, blue curve). To load ATP, CDK4 favors cyclin-D binding, which shifts the equilibrium toward the open ATP-binding site (cyclin-D/CDK4^Apo). Our findings also suggest that in the monomeric state, CDK2^ATP has a statistically significant higher, albeit small, affinity difference for ATP compared to CDK4^ATP (Fig. S2). This observation aligns with the fact that a crystal structure of monomeric CDK2 with ATP is available, while no such crystal structure is currently available for CDK4. Collectively, this supports higher CDK2 substrate phosphorylation efficiency than CDK4, indicating that CDK2 more readily shifts its conformation toward its active state and stabilizes the ATP. We attribute these observations to the structural and sequence differences, including the longer CDK4 loop^β3-aC and loop^β4-β5 (Fig. 4), which also interact with each other, preventing the αC-helix from moving inward. In line with our observations, in the case of B-Raf, EGFR, and HER2 kinases, a shortened loop between the β3-strand and the αC-helix leads to constitutive activation.⁴⁸

Figure 3. Conformational flexibility of CDK2 decreases on ATP-loading and cyclin-binding in contrast to CDK4.

(a) Superimposed snapshots show the cartoon representations of CDK2^Apo and cyclin-E/CDK2^ATP and that the conformational flexibility of CDK2^Apo is higher than that of cyclin-E/CDK2^ATP. The potential of mean force (PMF), ΔG_PMF, in the third column representing the relative free energy profile based on the calculation of the probability distributions for two atom pair distances, Lys-Glu and ValLeu, shows the conformational shifts of CDK2 from its apo state (CDK2^Apo) to the ATP- and cyclin-bound inactive state (cyclin-D/CDK4^ATP). Specifically, we use the distance between V18 in the roof of the active-site cleft and L134 in the bottom of the active-site cleft, and the distance between K33 on the β3-strand and E51 on the αC-helix of CDK2. (b) The same for the CDK4 systems with the PMF and two reaction coordinates. Note that the Val-Leu residue distance is evaluated by their Cα atoms, while the Lys-Glu distance is calculated using the distance between the nitrogen atom on the side chain of Lys and two oxygen atoms on Glue side chains, whichever that result in a shorter distance. See also Figures S3 and S4.

Figure 4. The interactions of loop^β3-aC and loop^β4-β5 in CDK4 prevent αC-Helix inward movement.

The conformations of the loops connecting the β3-strand to the αC-helix and the β4-strand to the β5-strand for CDK2^Apo (left panel) and CDK4^Apo (right panel). These two loops in CDK4 are longer than those in CDK2 (bottom panel), showing extensive interactions and blocking the αC-helix from moving inward.

Protein kinase activation involves conformational changes of both the αC-helix and the A-loop. Thus, inspection of the dynamics of the A-loop is also essential for monitoring the tendency of kinase conformational transitions. The coupling of the A-loop and the αC-helix in CDK2 and CDK4 shows a contrasting behavior. To illustrate how ATP loading and cyclin binding affect the shift towards active state, we examined the conformational changes of the A-loops, focusing on their coupling with the αC-helix (Fig. S5). We calculated the distance between the αC-helix and the αL12-helix of each CDK. αL12 is a small helix in the A-loop. In the inactive state it is spatially adjacent to the αC-helix (Fig. S5). In the CDK2 systems, it moves away from the αC-helix as the αC-helix moves inwards upon ATP loading (CDK2^ATP) or cyclin-E binding (cyclin-E/CDK2^Apo) (Fig. 5a). The αL12-helix moves farther away from the αC-helix in the cyclin-E/CDK2^ATP complex, suggesting that the A-loop tends to extend upon both ATP loading and cyclin-E binding. The outward movement of the αL12-helix destabilizes this inhibitory helix, leading to an extended A-loop as in the active form of CDK2 (Fig. S5). We suspect that the movements of the A-loop and αC-helix are coupled in CDK2 for activation. For the CDK4 systems, however, the αL12-helix slightly moves toward the αC-helix as the αC-helix moves inwards upon ATP loading (CDK4^ATP) and cyclin-D binding (cyclin-D/CDK4^Apo) (Fig. 5b). The αL12-helix is closer to the αC-helix in the cyclin-D/CDK4^ATP complex. This contrasting CDK4 behavior can be explained by the αL12-helix of CDK2 having weaker hydrophobic interactions with the αC-helix, with A151 in the αL12-helix compared to the I164 of CDK4. Recent experiments using double electron-electron resonance (DEER) spectroscopy revealed that wild-type CDK2 can populate the ensembles of the A-loop OUT conformation, but CDK2 with the A151I mutation (CDK4-specific) causes the ensembles to shift in the absence of the A-loop OUT conformation.⁴⁵ CDK2 with ATP and cyclin-E favors the A-loop OUT conformation, which eliminates the inhibitory helix and facilitates full A-loop extension. In contrast, CDK4 with ATP and cyclin-D retains the inhibitory helix that interacts with the αC-helix, stabilizing the inactive state. Thus, we hypothesize that CDK2 has a more dynamic and flexible A-loop movement than CDK4, which confers a kinetic advantage for its activation.

Figure 5. The coupling between the A-loop conformation and the αC-helix is stronger in CDK2 than CDK4.

(a) Snapshots representing the superimposed conformations of the A-loop of the CDK2 systems (left panel). Structural alignment of the A-loop of the ATP-loaded CDK2 in complex with cyclin-E (red cartoon) with respect to apo CDK2 (blue cartoon) (middle panel). Probability distribution functions of the αC-αL12 distance for CDK2^Apo (blue), CDK2^ATP (gray), cyclin-E/CDK2^Apo (green), and cyclin-E/CDK2^ATP (red) (right panel). (b) The same for the CDK4 systems (left and middle panels). The probability distribution functions show the αC-αL12 distance for CDK4^Apo (blue), CDK4^ATP (gray), cyclin-D/CDK4^Apo (green), and cyclin-D/CDK4^ATP (red) (right panel). Yellow arrows highlight the direction of the movement of αL12-helix. See also Figure S5.

Differential activation mechanisms of cyclin-E/CDK2 and cyclin-D/CDK4 via different binding preferences for ATP and cyclin partners

We decipher the activation mechanisms of cyclin-E/CDK2 and cyclin-D/CDK4, here aiming to elucidate the preferred order in which these kinases bind to their respective cyclin partners and ATP molecules. Above, we presented the conformational changes of the ATP-binding pocket upon cyclin binding to see how the cyclin stabilizes the pocket for efficient phosphorylation. For CDK2, ATP loading onto CDK2^Apo is highly favored due to the open pocket. However, cyclin-E binding to CDK2^Apo causes the pocket to close, reducing the probability of ATP loading. Thus, the most populated pathway for CDK2 activation involves ATP loading as the initial step. Binding of cyclin-D to CDK4^Apo results in the opening of the ATP-binding site, promoting subsequent ATP loading. However, ATP loading on CDK4^Apo is less likely due to the closed ATP-binding pocket. Thus, the most populated pathway for CDK4 is to first bind cyclin-D. Taken together, we suggest that inactive CDK2 preferentially loads ATP first and then binds cyclin-E (Fig. 6). Conversely, inactive CDK4 prefers to bind to cyclin-D first, which opens the ATP-binding pocket and facilitates ATP loading. In both cases, the ATP-bound inactive cyclin/CDK complexes undergo significant conformational changes, such as the αC-helix adopting an IN position and the A-loop extending outward to interact with their cyclin partners.

Figure 6. A schematic diagram illustrating the activation mechanism of cyclin-E/CDK2 and cyclin-D/CDK4.

Cyclins and CDKs in the cell cycle (left panel). Cell cycle progression is regulated by complexes of CDKs and their cyclin partners. The key regulatory steps and protein interactions contributing to the activations of CDK2 and CDK4 (right panel). The ATP-binding pocket of CDK2^Apo is open and flexible, facilitating ATP loading. Upon ATP loading, the pocket is closed, and cyclin-E binding promotes the active kinase conformation. In contrast, the ATP-binding pocket of CDK4^Apo is closed and rigid, impeding ATP entrance to the pocket. Cyclin-D binding induces the open conformation of the pocket, allowing ATP to load. Orange tubes are A-loops; cylinders are αC-helix; red dots are the phosphorylation sites on the A-loop. The major activation pathway for each CDK is indicated by bold arrows.

DISCUSSION

Cyclin-D/CDK4 and cyclin-E/CDK2 play distinct roles in the cell cycle. Here, we ask why cyclin-D/CDK4 preferentially acts in the G1 phase and cyclin-E/CDK2 in the G1/S transition. We also ask how they were optimized for their distinct functions. Our observations lead us to postulate that cyclin-D/CDK4 preferentially executes the G1 phase and cyclin-E/CDK2 the G1/S transition because cyclin-E/CDK2 has a faster activation time than cyclin-D/CDK4, and we detail how. Our innovative concept is based on observations from our data, cell biology, and experimental structural work.

The distinct mechanisms of activation of cyclin-D/CDK4 and cyclin-E/CDK2

In our scenarios, CDK2 is pre-organized to allow ATP loading prior to cyclin-E binding. In contrast, CDK4 exhibits a preference for cyclin-D binding before ATP loading. In line with this, (i) upon ATP loading and cyclin binding, the αC-helix of CDK2 experiences a significant inward movement, toward the active state. In contrast, the αC-helix of CDK4 remains relatively static in the inactive form, regardless of ATP loading or cyclin binding, suggesting a slower conformational change. (ii) The A-loop and the αC-helix of CDK2 are coupled, meaning that the inward movement of αC is correlated with A-loop extension. However, in CDK4, the A-loop and the αC-helix are more rigid and resist movement, requiring more energy to switch between states. (iii) CDK2 has a more flexible and responsive ATP-binding site than CDK4, providing a kinetic advantage for its activation. Significantly, (iv) in the cell, CDK2 activation is part of a positive feedback loop, resulting in switch-like behavior, while CDK4 activation is a more gradual process. In the G1 phase, active CDK4 (and CDK6) in complex with cyclin-D phosphorylates the Rb protein, triggering a partial release of the E2F transcription factor, which in turn induces cyclin-E expression. Cyclin-E activates CDK2, which further phosphorylates Rb, leading to the complete release of E2F, thereby enhancing cyclin-E expression further activating CDK2, and promoting hyperphosphorylation of Rb.^1,34 This canonical pathway demonstrates that CDK2 activation during the G1 phase to G1/S transition is part of a positive feedback loop. Recent studies proposed that the cyclin-D/CDK4 complex can either directly hyperphosphorylate Rb, leading to cell cycle progression into the S phase^49,50, or indirectly activate CDK2 by sequestering KIP/CIP family protein inhibitors away from CDK2, allowing CDK2 to hyperphosphorylate Rb and promote the cell’s entry into the S phase. Further, significantly, (v) in terms of duration, CDK4 primarily governs progression through the G1 phase, which is the longest phase of the cell cycle, lasting approximately 11 hours in rapidly proliferating human cells. In contrast, CDK2 regulates the G1/S phase transition, a relatively short period during which the cell decides whether to commit to DNA replication or not, as there is no turning back. Consequently, the activation time of CDK2 is shorter than that of CDK4, further highlighting the dynamic and time-sensitive nature of this critical phase in the cell cycle. In addition, (vi) related to the dynamics of the cell cycle, are the concentrations of cyclins in the multiple phases of the cell cycle, which have been assembled from multiple sources.⁵¹

Experimental support for the proposed mechanisms

Experimental structural reports point to (vii) differences in the crystal structures of cyclin-E/CDK2 and cyclin-D/CDK4. They indicate that cyclin-E/CDK2 has a more stable active conformation. Proteins can adopt various conformations, and their crystal or cryo-EM structures only capture snapshots of these.^52–55 The most likely scenario is that the conformation captured in the crystal structure represents the one with the lowest energy under the crystallization conditions. A comparison of the available crystal structures of CDK2 and CDK4 provides evidence for differences in their activation dynamics. CDK2 has been observed in the active state (αC-in and A-loop extended) when complexed with cyclin-E.^36,56 In contrast, a crystal structure of fully active cyclin-D/CDK4 complex was not available until late 2022 when it was published by Gharbi et al.³⁷ This observation suggests that CDK2 has a more stable active conformation, implying faster and more efficient activation. The higher stability of the active CDK2 conformation in complex with cyclin-E, as opposed to CDK4, may reflect a lower energy barrier for the conformational changes required for activation. This lower energy barrier and faster activation kinetics would allow CDK2 to transition more rapidly from an inactive to an active state during the critical “point of no return” for committing to the DNA replication S phase. This difference in activation dynamics could have functional implications for the regulation of cell cycle progression and response to extracellular signals. Thus, the comparison of the crystal structures when considering their associated conformational potential energy landscapes supports the argument that cyclin-E/CDK2 is faster to activate than cyclin-D/CDK4. Lastly, the structural differences in the phosphorylated cyclin-E/CDK2 and cyclin-D/CDK4 complexes on the A-loop also suggest that the CDK2 complex activates faster than the CDK4 complex. Upon binding cyclin-E, CDK2 adopts an active conformation, in which the active site on the A-loop is buried in the complex.^56,57 This intrinsic conformational stability also hinders phosphatases from accessing the active site and dephosphorylating CDK2. In contrast, when bound to cyclin-D, CDK4 exhibits an exposed active site on the A-loop,^36,56 which facilitates fine-tuning of CDK4 activation by growth factors signaling.⁵⁷ Although this dynamic regulation provides a higher degree of adaptability, it may result in slower activation due to the active state being continually modulated by phosphorylation and dephosphorylation.⁵⁸

Most importantly, from the standpoint of the cell, two recent studies used single-cell experiments with live-cell reporters for CDK4/6 and CDK2, to track their activities as a function of time since mitogen release.^35,50 We observe that in both works, for the aggregated cell data, the rate at which CDK4 and CDK2 activities increase is faster for CDK2 than for CDK4. This rate is described by the slope of the curve during CDK4 and CDK2 activities. These experimental findings further corroborate our cell cycle control and sequential activation perspective, where CDK4/6 activation is carefully coordinated prior to CDK2 activation.

Our proposed mechanisms can explain why cyclin-E/CDK2 works at the sharp transition between G1 and S while cyclin-D/CDK4 is active during the G1 phase

Here we perform simulations of the CDK2 and CDK4 (starting in their OFF states) with and without the appropriate cyclins in the presence and absence of ATP to investigate the intrinsic conformational propensities of each kinase. During the simulations, we observe that (i) CDK2 is structurally more poised to bind ATP due to flexible A-loop and (ii) more open nucleotide binding cleft as compared to CDK4. We also see that (iii) CDK2 more readily switches into the ON state (as indicated by the tendency of αC-helix position) upon nucleotide and/or cyclin binding as compared to CDK4. Lastly, (iv) we observe a more dynamic αL12-helix in the CDK2 which may allow for easier switching to the ON state of CDK2 as compared to CDK4. Taken together, CDK2 more readily binds ATP and is more readily switched into the ON conformation than CDK4. These observations lead us to suggest that these differences may help explain why CDK2 works at the sharp transition between G1 and S while CDK4 is active during the G1 phase. These observations need however to be put in the methodological and broader cellular context.

Considerations of the timescales and the biology

Conformational transitions in kinases can happen at the millisecond and slower timescales, which are not captured in microsecond simulations. The timescale of the transitions observed here, and cell cycle transitions, are orders of magnitude different⁵⁹, and the biological picture is fundamentally complex. Both kinases interact with CIP and Ink proteins, and recent work showed that the phosphorylation state of CIPs play an important role in CDK4 activation²⁶. When interacting with the molecular chaperone Hsp90/Cdc37, CDK4 populates a re-arranged conformation⁶⁰, much beyond that observed in the simulations. Thus, the biological timing and function of CDK2 and CDK4 appear largely driven by deep energy wells created through interactions with cellular proteins and chaperones. Hsp90 can efficiently modulate the allosteric interactions and long-range communications for client protein activation. At the same time, as we have shown here, the energetic fluctuations captured in the simulations are critical at the protein level, beyond that in the cellular environment.

Finally, the major contributor to fast CDK2 activation is the feedback mechanism between CDK2 and p21. A relatively small number of active CDK2 molecules can rapidly trigger p21 degradation releasing a burst of already active kinase molecules at the G1/S transition.⁶¹

Implications to drug discovery

Our studies offer insights into the conformational changes involved in cyclin-E/CDK2 and cyclin-D/CDK4 activation and their implications for drug discovery, particularly for CDK4, which is often targeted in cancer. CDK4 is overactive in cancer cells^62,63, and has a role in cell cycle (dys)regulation⁶⁴, making it an attractive drug target. Our findings of the distinct activation mechanisms of cyclin-E/CDK2 and cyclin-D/CDK4 suggest potential strategies for designing selective allosteric inhibitors for CDK4. The gradual activation process of CDK4 provides a longer time frame for inhibitors to bind, in contrast to CDK2. One possible approach for CDK4 allosteric inhibitors is stabilizing the inactive conformation, a classic allosteric kinase inhibitor design principle. Design strategies could also involve targeting conformational changes in CDK4 activation or blocking cyclin-D binding to CDK4. Allosteric inhibitors offer an alternative to ATP-competitive inhibitors, as they can be less toxic and more selective. Recent developments in allosteric inhibition of CDK2, targeting the interface between CDK2 and its cyclin partner⁶⁵, provide promising examples for CDK4 inhibitor design.⁶⁶ Accounting for the dynamic nature of protein binding sites and ATP pocket geometry is crucial for effective kinase inhibitor development. As to orthosteric drugs, as the recent crystal structure of CDK11 bound to the selective inhibitor OTS964 showed, evolutionary variations in the kinase domain can be exploited for drugs for closely related kinases.⁶⁷

CONCLUSIONS

The cell cycle plays a cardinal role in cell life and death. Under normal physiological conditions, it is carefully orchestrated; when dysregulated, it can lead to uncontrolled proliferation and cancer, to oncogene-induced senescence (OIS), as well as aberrant differentiation in neurodevelopmental disorders.⁶⁸ Cell cycle regulation entails complex processes. Full understanding requires grasp of its cellular and its structural biology. Several CDKs (e.g., CDK1, CDK2, CDK4 and 6) and cyclins (e.g., A, B, D, E) are involved in these processes. Despite their structural similarities and functional overlap, specific complexes are favored in distinct cell cycle stages: cyclin-D/CDK4/6 in the G1, cyclin-E/CDK2 in the G1/S transition; cyclin-A/CDK2 in the S stage; cyclin-A/CDK1 in the G2; and cyclin-B/CDK1 in the M stage. Our challenging aim is to figure out the structural basis of the cyclin/CDK preferences for the distinct cell cycle stages and merge these with cell data on protein function and regulation. We also aim to determine key structural characteristics that influence the catalytic efficiencies of these CDK complexes. To understand the workings of the cell these should be integrated with the structural changes of the protein complexes and the biology.

Within this framework, here we connected the mechanisms of activation of cyclin-D/CDK4 and cyclin-E/CDK2 to their functions in the G1 phase and G1/S transition in the cell cycle. These led us to postulate slower activation of cyclin-D/CDK4 and faster activation of cyclin-E/CDK2. We observed that cyclin-E/CDK2 prefers ATP loading prior to cyclin-E binding, while CDK4 prefers cyclin-D binding before ATP loading. We further observed that cyclin-E/CDK2 has a faster activation time than cyclin-D/CDK4 due to the conformational changes involved in its activation pathway, its flexible and responsive ATP-binding site, and the readily accessible stable active conformation.

We also highlighted the importance of considering the conformational energy landscape for understanding how the activation mechanisms of cyclin-E/CDK2 and cyclin-D/CDK4 integrate with cell biology to accomplish their functional roles. Overall, these innovative mechanistic findings decipher overlooked hallmarks of the regulation of cell cycle progression and may inform drug development. To our knowledge, our work pioneers the connection between the cyclins-CDKs mechanisms and their roles in cell cycle function.

Our study suggests that (i) the activation dynamics of CDK2 can be incorporated into a cellular expression level-based model of CDK2 activation, and that (ii) the mechanisms of CDK4 and CDK2 are distinct as well. Notably, the mechanism of activation of CDK1 and CDK2 are known to be distinct.⁵⁸ Finally, (iii) our work raises the question of whether observations made here for CDK4 apply to CDK6, where different activation mechanisms have been proposed.⁶⁹

Our work provides an unprecedented mechanistic understanding of the distinct activation scenarios of cyclin-D/CDK4 and cyclin-E/CDK2 in cell cycle regulation, underpinning the slower activation of cyclin-D/CDK4 in the more extended G1 phase and the rapid activation of cyclin-E/CDK2 in the brief G1/S transition. Leveraging a range of experimental data and molecular dynamics simulations, we elucidate the inherent conformational dynamics and activation pathways of these key cell cycle regulators. Our findings not only address a long-standing question in cell cycle biology but also the design of more targeted and effective inhibitors against CDK4, opening new venues in cancer treatment.

STAR★Methods

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Ruth Nussinov (NussinoR@mail.nih.gov).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and study participant details

No experimental models were used in this work.

Method details

Modeling of inactive CDKs and CDKs complexes

We modeled the inactive conformations of CDK2 and CDK4 in their monomeric and cyclin-bound states in the presence and absence of ATP. The initial coordinates for the inactive CDK2 monomers were taken from the crystal structures of the ATP-bound state(PDB ID: 1HCK) and the apo state (PDB ID: 1HCL), which generate monomeric inactive CDK2^ATP and CDK2^Apo, respectively. The inactive cyclin-E/CDK2 complexes were modeled by replacing the active CDK2 in the cyclin-E/CDK2 complex (PDB ID: 1W98) with inactive CDK2. The same CDK2 crystal structure from 1HCL, the inactive monomeric apo-CDK2, was used to model the monomeric (CDK2^Apo) and complex (cyclin-E/CDK2^Apo) systems. Similarly, the same crystal structure of inactive ATP-bound CDK2 from 1HCK was used to model the monomeric (CDK2^ATP) and complex (cyclin-E/CDK2^ATP) systems. For such short missing residues 37–40 in 1HCK, the coordinates were constructed as a loop using the internal coordinate tables in the CHARMM program. For the starting structures of CDK2, 1HCL and 1HCK were taken from the same study and are very similar, including residue conformations in the ATP cleft with the RMSD of 0.39 Å between these two structures.⁷⁰ To confirm that the two very similar starting crystal structures (1HCL and 1HCK) used in the MD simulation of CDK2 are not the source of differences in the conformational dynamics, but rather that these differences are due to ATP loading and cyclin binding, we performed an additional simulation of CDK2 complex detailed in Fig. S6. The results confirmed this hypothesis. The missing residues 81–87 on chain B of 1W98 correspond to an unstructured N-terminal tail of cyclin-E, so we ignored this region in the cyclin-E model, which yields the N-terminus starting from residue 88. For CDK4 systems, the initial coordinates for the inactive CDK4 monomers were extracted from the crystal structure of the inactive cyclin-D/CDK4 complex (PDB ID: 2W9Z)³⁶. We modeled ATP into the inactive nucleotide-free CDK4 based on the ATP-bound crystal structure of CDK2 in 1HCK, generating the monomeric (CDK4^ATP) and complex (cyclin-D/CDK4^ATP) systems. We simulated all systems in the presence and absence of ATP in the active site. We performed a series of minimization and dynamics cycles using the CHARMM program to eliminate any steric clash or unfavorable interactions, and meticulously checked the optimized structure and interaction energy to ensure that there are no unphysical interactions and frustrated structures. The simulations are listed in Table S1, and snapshots of all systems are in Fig. S1. We used the TIP3P water model and added Na⁺ and Cl⁻ to neutralize the solvated systems and maintain a physiological salt concentration of 150 mM.

Atomistic molecular dynamics simulation protocol

We performed molecular dynamics (MD) simulations using the protocol established in our previous publications.^75–79 The simulations consisted of several stages. We began by conducting 10,000-step energy minimizations using the conjugate gradient minimization method to construct the systems. This step was crucial for eliminating unfavorable contacts between atoms in the systems. For each system, we performed a 2 μs all-atom explicit-solvent MD simulation under the NPT ensemble (constant number of atoms, pressure, and temperature) and 3D periodic boundary conditions. The NAMD 2.14 package⁷² and CHARMM⁸⁰ all-atom force field (version 36m)⁸¹ were employed in these simulations. The pressure and temperature were maintained at 1 atm and 310 K using the Langevin piston control algorithm and the Langevin thermostat method, respectively. A damping coefficient of 1 ps⁻¹ was applied. All covalent bonds involving hydrogen atoms were constrained using the RATTLE method. This allowed us to employ the velocity Verlet algorithm for integrating the Newtonian motion equation with a larger time step of 2 fs. The interaction potentials between atoms were computed using the particle mesh Ewald (PME) method for long-range electrostatic interactions (grid spacing of 1.0 Å) and switching functions for short-range van der Waals (vdW) interactions (twin-range cutoff at 12 and 14 Å).

Quantification and statistical analysis

We used PyMOL⁸² and VMD⁷³ software to visualize and analyze the molecular trajectories. The CHARMM and VMD packages⁷³ were utilized for result analysis, employing FORTRAN and TCL scripts. The root-mean-square deviation (RMSD) profiles indicated that all studied systems reached convergence after 500 ns. To ensure reproducibility in our molecular dynamics simulations, we followed our standard simulation protocol and conducted at least two independent simulations for each system. Across these replicated simulations, consistent and comparable outcomes were observed. These assessments are based on qualitative assessments and visual inspections of the simulation trajectories since our primary focus was on exploring qualitative aspects of protein dynamics. For statistical robustness, the data were aggregated by computing the average from all replicates. Furthermore, averages were taken over the last 1 μs trajectories to ensure data reliability and consistency.

Key resources table

REAGENT OR RESOURCE	SOURCE	IDENTIFIER
Deposited data
CDK2Apo	SchulzeGahmen et al.70	PDB:1HCL
CDK2ATP	SchulzeGahmen et al.70	PDB:1HCK
Cyclin-E/CDK2Apo (pT160)	Honda et al.71	PDB:1W98
Cyclin-D/CDK4Apo	Day et al.36	PDB:2W9Z
Software and algorithms
NAMD 2.14	Phillips et al.72	http://www.ks.uiuc.edu/Research/namd/
VMD	Humphrey el al.73	http://www.ks.uiuc.edu/Research/vmd/
Pymol	DeLano et al.74	http://www.pymol.org

Highlights:

• Cyclin-D/CDK4 activation is slower than cyclin-E/CDK2, informing cancer drug design

• There are distinct activation speeds for G1 phase and G1/S transition

• Conformational dynamics of CDKs inform novel therapeutic strategies

• Study bridges cell cycle regulation with CDK activation mechanisms