Oana-Maria et al. Frontiers in Pharmacology, 12 (2021) https://www.frontiersin.org/article/...ar.2021.757120. DOI=10.3389/fphar.2021.757120 . Creative Commons Attribution License (CC BY).
Cyclin-dependent kinases (CDKs) are key players in cell cycle regulation. So far, more than ten CDKs have been described. Their direct interaction with cyclins allow progression through G1 phase, transitions to S and G2 phase and finally through mitosis (M). While CDK activation is important in cell renewal, its aberrant expression can lead to the development of malignant tumor cells. Dysregulations in CDK pathways are often encountered in various types of cancer, including all gastrointestinal (GI) tract tumors. This prompted the development of CDK inhibitors as novel therapies for cancer. Currently, CDK inhibitors such as CDK4/6 inhibitors are used in pre-clinical studies for cancer treatment. In this review, we will focus on the therapeutic role of various CDK inhibitors in colorectal cancer, with a special focus on the CDK4/6 inhibitors.
Cyclin-Dependent Kinases and Their Role in Cell Cycle Progression
Cell cycle is defined as the process through which the cell replicates all its genomic material and divides into two identical cells (Alberts et al., 2002). It consists of four phases: gap 1 (G1), where the cell grows in size and transcribes the RNA and protein necessary during cell division; synthesis or S phase, where all chromosomes are being replicated; gap 2 (G2), where cell growth and protein synthesis continue; and mitosis or M phase, where the cell restructures its membrane and organizes the newly synthesized chromosomes and then divides into two daughter cells. Before entering cell cycle, highly proliferative cells such as stem cells and lymphocytes are in a reversible cell cycle arrest, known as quiescence or gap 0 (G0). However, other cells such as neurons or adipocytes are irreversibly arrested in G0 phase, a phenomenon often described as cellular senescence. Senescence is also predominant in highly damaged cells, acting as a protective mechanism during the DNA damage response (DDR) (Terzi et al., 2016).
Each cell cycle phase, as well as transitions from one phase to the other, are tightly regulated by interactions between cyclins and cyclin-dependent kinases (CDKs) (Johnson and Walker, 1999). In general, cyclins directly bind CDKs and induce the formation of cyclin—CDK complexes. This promotes CDK activity and therefore ensures activation of specific transcriptional programs that allow cell cycle progression. More than ten CDKs are known to be involved in various events during cell cycle. From these, CDK1, 2, 3, 4, and 6 directly mediate cell cycle progression.
Transition from quiescence or G0 phase in G1 phase is modulated by growth factor signals or mitogenic stimulation. These result in the upregulation of Cyclin D, which binds to and activates CDK4 and CDK6 to promote cell commitment to enter G1 phase (Jinno et al., 1999; Lea et al., 2003). High CDK4/6 expression and activation ensures cell progression through G1 phase (Mende et al., 2015; Topacio et al., 2019).
On the molecular level, CDK4 and 6 phosphorylate Retinoblastoma (Rb) and promote the accumulation of E2F, a direct regulator of genes necessary during DNA synthesis. Furthermore, CDK4 and CDK6 activation initiates cell growth through activation of mammalian target of rapamycin complex 1 (mTORC1) (Romero-Pozuelo et al., 2020). Besides, CDK4 and 6 are involved in the control of DNA replication mechanisms (Braden et al., 2008). Along with CDK4/6, CDK2 and CDK3 are also activated during G1 phase. Rb phosphorylation, and therefore the accumulation of E2F during G1 phase, directly mediate the upregulation of Cyclin E in late G1 phase, which binds and activates CDK2. Formation of CDK2/Cyclin E complex maintains Rb phosphorylated in order to promote G1/S phase transition (Massague 2004; Horiuchi et al., 2012). However, CDK3 upregulation during late G1 phase seems to be independent of Cyclin D, E or A binding (Braun et al., 1998). Interestingly, the upregulation of CDK2 has been also shown to be important during the G1/S checkpoint in response to DNA damage. For example, knocking-down CDK2 in the HCT116 tumor cell line significantly reduced p53 phosphorylation in response to hydroxyurea (HU) and suppressed G1/S cell cycle arrest (Bacevic et al., 2017). Some recent studies also described a role of CDK2 directly after mitosis, as an intermediate level will remain in the cells that continue proliferating, while those that lack CDK2 can enter quiescence or so called gap 0 (G0) (Spencer et al., 2013; Gookin et al., 2017). On the other hand, high levels of Cyclin C/CDK3 have been reported to directly mediate quiescence (Ren and Rollins 2004).
The beginning of S phase is marked by increasing levels of Cyclin A, which binds CDK2. The complex formed by Cyclin A/CDK2 drives the cells through S phase and promotes DNA replication. During late S/G2 phase, increased levels of Cyclin A induce CDK1 activation, which drives entry into mitosis (Gavet and Pines, 2010; Kalous et al., 2020). Later, the formation of CDK1/Cyclin B complex triggers progression through M phase. Along with its important role in successful cell mitosis (Vassilev et al., 2006), CDK1 can also influence the remodeling of cell adhesion complexes during G1, S and G2 cell cycle phases (Jones et al., 2018) and promotes protein synthesis during proliferation (Haneke et al., 2020). Interestingly, CDK1 is reported to be the only necessary cyclin-dependent kinase during cell cycle, being able to bind to all cyclins and drive all events during cell division (Santamaria et al., 2007).
Several other CDKs are known to be involved in cell cycle progression as well. CDK7, for example, is an important cell cycle regulator. Its binding to Cyclin H and mating-type 1 protein (Mat1) induces the formation of CDK-activating kinase (CAK) complex. CAK activity is crucial to promote CDK2 and CDK1 binding to cyclins, therefore allowing cell division (Fisher and Morgan, 1994; Larochelle et al., 2007; Olson et al., 2019). CDK5 upregulation is mostly observed in, but not limited to, neurons, and is often correlated to cell apoptosis. Nevertheless, it can also regulate the cell cycle by phosphorylating Rb and interacting with E2F during G1 phase (Zhang et al., 2010; Chang et al., 2012; Futatsugi et al., 2012). CDK8 is a partner of Cyclin C and its expression has been shown to be important in stabilizing Cyclin C activity during cell cycle (Tassan et al., 1995; Barette et al., 2001). Interestingly, CDK8 and Cyclin C, as well as CDK19/Cyclin C complex, are strongly required during p53-dependent p21 transcriptional activation, for cell cycle arrest in response to DNA damage (Donner et al., 2007; Audetat et al., 2017). Last, cyclin-dependent kinases such as CDK9 and CDK13 are not directly controlling cell cycle phase transitions, but are rather involved in transcription mechanisms, by associating with Cyclin T or Cyclin K (Garriga et al., 2003; Yu et al., 2010; Greifenberg et al., 2016).
To summarize, entry into cell cycle depends on mitogenic or growth factor signals. CDK4/6/Cyclin D complex formation promotes Rb phosphorylation and accumulation of free E2F, which ensures progression through G1 phase. CDK5 activity also increases E2F levels during G1. High levels of E2F during late G1 induce CDK2/Cyclin E complex that in return further phosphorylates Rb and promotes G1/S transition. At the beginning of S phase, Cyclin E levels decrease and CDK2 forms a complex with the increasing Cyclin A, which not only ensures progression through S phase, but also transition into G2 phase. CDK2/Cyclin A complex is especially regulated by the CDK7/Cyclin H/Mat1 complex, also described as CAK. CAK also regulates CDK1/Cyclin A complex formation during late G2 and Cyclin B binding to CDK1 during mitosis. Any disturbances to the cell cycle machinery will result in cell cycle arrest. CDK2 and CDK3 are especially important in mediating either quiescence or senescence. Indirectly, CDK8, 9, 13, and 19 also mediate cell cycle, being involved in the transcription machinery, while CDK5 can directly modulate apoptosis as well. A schematic representation of the important role of CDKs in cell cycle is shown in Figure 1. While normal cells are able to activate the necessary mechanisms for cell cycle arrest when the DNA is damaged, these pathways are usually suppressed or non-existent in tumor cells, enabling them to continue progression through cell cycle. The following sections will address the CDK’s role in the tumor cell division and how therapies targeting CDKs can modulate CRC development.
Aleem and Arceci . Targeting cell cycle regulators in hematologic malignancies. Article in Frontiers in Cell and Developmental Biology 2015. DOI: 10.3389/fcell.2015.00016. Creative Commons Attribution 4.0 International
In order for a cell to undergo successful division, it has to perform four key tasks in a highly ordered fashion. First, there is a prepara- tory synthetic phase (G1) that results in an increased cell size in anticipation of DNA replication (S phase). Cells then proceed through (G2-phase) to prepare to equally segregate duplicated DNA (M phase) and finally divide into two equal daughter cells. From G1 a cell can also exit the cell cycle and enter a state of qui- escence (G0), undergo differentiation or re-enter the cell cycle to proliferate in response to mitogenic signals.
The core molecular machinery controlling the mammalian cell cycle consists of a family of serine/threonine protein kinases called cyclin-dependent kinases (CDKs). These are catalytic subunits, which are activated in most cases by association with cyclin regulatory subunits. The activity of CDK/cyclin complexes is further regulated by CDK-inhibitors (CKIs), phosphorylation and dephosphorylation, ubiquitin-mediated degradation, transcriptional regulation, substrate recognition, and subcellular localization (Aleem and Kaldis, 2006). The family of CDKs/cyclins/CKIs contains more than 30 members (Figure 1; Supplementary Table 1) and they are implicated in essential cellular functions such as transcription, DNA damage repair, epigenetic regulation, metabolism, proteolytic degrada- tion, stem cell self renewal, neuronal functions, and spermato- genesis (Lim and Kaldis, 2013).
FIGURE 1 | Cyclin-dependent kinases (CDKs) and their cyclin regulatory subunits. CDK-cyclin complexes with direct functions in regulating the cell cycle. CDK3/cyclin C drives cell cycle entry from G0. CDK4/6/cyclin D complexes initiate phosphorylation of the retinoblastoma protein (pRb) and they sequester p21Cip1 and p27kip1 (not shown), which are both inhibitors of CDK2, thus promoting the activation of CDK2/cyclin E complex. In late G1, CDK2/cyclin E complex completes phosphorylation and inactivation of pRb, which releases the E2F transcription factors and G1/S transition takes place. DNA replication takes place in S phase. CDK2/cyclin A complex regulates progression through S phase and CDK1/cyclin A complex through G2 phase in preparation for mitosis (M). Mitosis is initiated by CDK1/cyclin B complex. The activity of CDK1/cyclin B is tightly regulated by activating phosphorylation by the CDK-activating kinase CAK (a heterodimer of cyclin H-CDK7-MAT1) and inhibitory phosphorylations by Wee1 and Myt1 on Tyr15 and Thr14 (not shown). The specific CDK4/CDK6 pharmacological inhibitors described in this study are shown.
CDKs with Direct Functions in Cell Cycle Regulation
The classical CDKs that directly regulate the mammalian cell cycle in complexes with cyclin subunits include CDK3, CDK4, CDK6, CDK2, and CDK1. CDK3 promotes cell cycle entry from quiescence in association with cyclin C (Ren and Rollins, 2004). CDK8 has also been suggested to play a role in cell cycle entry from G0 and in the G1/S transition (Szilagyi and Gustafsson, 2013). In its simplest model, the mammalian cell cycle pro- ceeds as follows: In early G1, CDK4/CDK6 in complex with cyclin D receive mitogenic signals that result in activation of cell cycle entry (Figure 1). Key signaling events include initia- tion of retinoblastoma protein (pRb) phosphorylation and the sequestration of p21Cip1 and p27kip1, which are both inhibitors of CDK2, thus promoting the activation of CDK2/ cyclin E com- plex (Sherr and Roberts, 1999). In late G1, CDK2 in complex with cyclin E completes the phosphorylation and hence inactivation of pRb, which in turn releases the E2F transcription factors (Kaldis and Aleem, 2005). E2F promotes transcription of cyclin E that is necessary for the G1/S transition.
Progression through S phase is mediated by CDK2/cyclin A complex. Mitosis is then initiated by CDK1/cyclin B com- plexes. CDK1/cyclin A complexes contribute to the preparation for mitosis in G2 phase (Nigg, 1995; Edgar and Lehner, 1996). The activity of CDK1/cyclin B is tightly regulated by activating phosphorylation by the CDK-activating kinase (CAK) (a het- erodimer of cyclin H and CDK7) and inhibitory phosphoryla- tions by WEE-1 and Myt1 on Tyr15 and Thr14. Mitosis starts after WEE-1 is degraded and CDC25C phosphatase releases the inhibitory phosphorylation on CDK1/cyclin B.
CDKs with Transcriptional and Other Functions
In addition to their direct role in the mitotic cell cycle regula- tion, some classical CDK/cyclin complexes have essential func- tions in meiosis, such as CDK2 (Berthet et al., 2003; Ortega et al., 2003), in transcription and/or DNA repair (e.g., CDK1, CDK2, CDK6) (Satyanarayana and Kaldis, 2009; Neganova et al., 2011; Wohlbold et al., 2012; Kollmann et al., 2013; Lim and Kaldis, 2013; Scheicher et al., 2015). Other CDKs (referred to in Supplementary Table 1 as non-classical CDKs) act by activat- ing the classical CDKs, such as CDK7/cyclin H (CAK) and the related CDK20, also known as cell cycle related kinase (CCRK) (Wohlbold et al., 2006). Some CDKs function mainly in influ- encing transcription by phosphorylating the carboxy-terminal domain (CTD) of ribonucleic acid (RNA) polymerase II (RNA pol II) (Bose et al., 2013; Lim and Kaldis, 2013) (Supplemen- tary Table 1). This phosphorylation also serves as a platform for RNA processing and chromatin regulation (Hirose and Ohkuma, 2007).
CDKs that have important transcriptional roles include CDK7/cyclin H/MAT1 complex, a component of the basal tran- scription factor, TFIIH and facilitates transcriptional initiation (Shiekhattar et al., 1995). CDK8/cyclin C, in addition to its role in transcription (Gonzalez et al., 2014), is also involved in the Wnt/β-catenin pathway (Firestein et al., 2008) and in inhibition of lipogenesis (Zhao et al., 2012). Cyclin C can recruit CDK8 or CDK19 to the CDK8 module of the Mediator complex, which can function as a positive or negative regulator of transcription by RNA pol II (Li et al., 2014; Trakala and Malumbres, 2014). CDK3/cyclin C also plays a role in NHEJ-mediated DNA damage repair (Tomashevski et al., 2010). While CDK9 in complex with cyclin T forms the phospho-transcription elongation factor b (p-TEFb) and promotes transcriptional elongation (Shapiro, 2006), CDK9 also functions in the DNA damage response when
complexed with cyclin K (Yu et al., 2010). CDK10/cyclin M phosphorylates the Ets2 transcription factor and positively con- trols its degradation by the proteasome. Ets2 plays key roles in cancer and development (Guen et al., 2013). CDK11/cyclin L controls the assembly of the RNA pol II mediator complex (Dro- gat et al., 2012). CDK12 and CDK13 in complex with cyclin K control RNA pol II transcription (Bartkowiak et al., 2010; Blazek et al., 2011; Cheng et al., 2012), and CDK12/cyclin K controls DNA damage response (Blazek et al., 2011). These functions are summarized in Supplementary Table 1.
CDKs with additional functions include CDK5, which plays an important role in controlling neural development and post- synaptic signal integration (Kim and Ryan, 2010), and influ- ences epigenetic regulation through its interaction with Dnmt1 (Lavoie and St-Pierre, 2011). CDK5 also regulates degranulation in human eosinophils (Odemuyiwa et al., 2015). CDK5 plays a role in diabetes; it phosphorylates the peroxisome proliferator- activated receptor γ (PPARγ) at Ser 273, thus stimulating dia- betogenic gene expression in adipose tissue (Choi et al., 2010). CDK5 also suppresses ERK2 through direct phosphorylation (Banks et al., 2015). CDK14/cyclin Y is involved in the Wnt/β- catenin pathway (Davidson et al., 2009). CDK15 was reported to attenuate Tumor necrosis factor-related apoptosis-inducing lig- and (TRAIL)-induced apoptosis by phosphorylation of survivin (Park et al., 2014). CDK16 in complex with cyclin Y regulates spermatogenesis (Mikolcevic et al., 2012). CDK17 and CDK18 are not well-characterized but a recent report demonstrated that CDK18 is phosphorylated and activated by cyclin A2 and cAMP- dependent protein kinase (PKA) and its knockdown induced polymerized actin accumulation in peripheral areas and cofilin phosphorylation (Matsuda et al., 2014). Furthermore, CDK18 is highly expressed in diabetic human pancreatic islets (Taneera et al., 2013). CDK20 plays a role in ciliogenesis (Ko et al., 2010; Phirke et al., 2011; Yang et al., 2013a)(Supplementary Table 1).
Because of the transcriptional role played by multiple CDKs, pan CDK inhibitors, such as flavopiridol, not only arrest the cell cycle but can interfere with cellular transcription and reduce the expression of short-lived mRNAs and proteins, which include cell cycle regulators and survival factors, such as Mcl-1, thus promoting cell death (Bruyere and Meijer, 2013).
Structural insights into the functional diversity of the CDK–cyclin family
Since their characterization as conserved modules that regulate progression through the eukaryotic cell cycle, cyclin-dependent protein kinases (CDKs) in higher eukaryotic cells are now also emerging as significant regulators of transcription, metabolism and cell differentiation. The cyclins, though originally characterized as CDK partners, also have CDK-independent roles that include the regulation of DNA damage repair and transcriptional programmes that direct cell differentiation, apoptosis and metabolic flux. This review compares the structures of the members of the CDK and cyclin families determined by X-ray crystallography, and considers what mechanistic insights they provide to guide functional studies and distinguish CDK- and cyclin-specific activities. Aberrant CDK activity is a hallmark of a number of diseases, and structural studies can provide important insights to identify novel routes to therapy.
Members of the cyclin-dependent protein kinase (CDK) family were originally characterized as serine/threonine-specific protein kinases activated by the expression of cyclin partners to drive the eukaryotic cell cycle . Within the CMGC branch of the kinome, 20 proteins are now considered to be members of the CDK family that can be grouped into different phylogenetic sub-branches (see  for criteria for inclusion, illustrated and updated in ). In overview, in addition to those CDKs that regulate the cell cycle (CDKs 1, 2, 4 and 6), a substantial sub-branch of the family (CDKs 7, 8, 9, 12 and 13) regulates transcription through phosphorylation of the heptad repeats that comprise the C-terminal tail of RNA polymerase II (CTD) . CDK7 is unusual in that it also indirectly regulates the cell cycle by activating CDKs 1, 2, 4 and 6 [5,6]. CDK3 phosphorylates retinoblastoma protein (pRB) to promote the transition from quiescence (G0) into G1 .
Other CDKs (CDKs 5, 10, 11, 14–18 and 20) have more diverse, CDK-unique functions that are frequently tissue-specific . For example, CDK5 was one of the first CDKs to be characterized in non-cycling cells . CDK10 is implicated in regulating gene transcription, but not through RNA pol II phosphorylation. It phosphorylates diverse substrates including the ETS2 oncoprotein and the protein kinase PKN2, and mutations in its cognate cyclin, cyclin M, result in STAR syndrome, a human developmental disorder [10,11]. CDK10 mutant and knockout mice also show growth and developmental delays . CDK11–cyclin L complexes regulate RNA splicing, studied, for example, in the context of human immunodeficiency virus (HIV) transcript processing . However, insights into these CDK–cyclin interactions are limited by the lack of structures for CDK10- and CDK11-containing complexes.
To partner the CDKs in humans, approximately 30 proteins are classified as cyclins [3,8]. The cyclins share very little sequence homology, but are structurally defined by the presence of either one or two copies of the cyclin box fold (CBF) [3,14]. The structures of monomeric CDK2 and cyclin A and of CDK2–cyclin A in various activation states were together taken to be a model for the regulation of the CDK family by cyclin binding and phosphorylation . However, subsequent studies have shown that even closely related CDKs have distinct structural and sequence peculiarities. These differences translate into diverse substrate preferences and modes of regulation. CDK activity is wired into cell-type-specific signalling networks with the result that, taken together, knockout mice studies reveal both the redundancy inherent within the cell cycle CDKs, but also their tissue-specific activities (, CDK1; [17,18], CDK2; [19,20], CDK4; and [21,22], CDK6).
Dysregulation of CDK activity, either through activation of proteins that promote CDK activity or inactivation of oncogene-induced senescence pathways, is a common occurrence in various cancers [23–27]. Identifying and characterizing those cancers that require specific CDK activities for proliferation will provide the mechanistic understanding to better employ CDK-selective inhibitors. However, the importance of CDK activity to cancer initiation, growth and differentiation is further complicated by the emerging cell-cycle-independent roles of individual CDKs and cyclins in mammalian cells that are, respectively, cyclin and CDK partner-independent [28–30].
In this review, we compare and contrast the various monomeric CDK, CDK–cyclin and CDK-containing assemblies for which structures have been determined, and discuss how they might help to elucidate the different mechanisms that regulate CDK activity. Proteomic studies are identifying multiple proteins that bind to CDKs and cyclins that apparently do not share sequence features with proteins for which structures bound to CDKs or cyclins are available (table 1). A comparison of the structures of CDK–cyclin complexes reveals how the CDK and cyclin partners can differ in their relative disposition and the alternative surfaces that can be exploited to recognize CDK substrates and regulators. The extent to which protein interaction sites are conserved and recycled within the CDK and cyclin families is yet to be fully explored, but will be reviewed here. The kinetic and catalytic mechanism of protein kinases including CDK2 was reviewed in 2012 . The structures of CDK–cyclin complexes bound to ATP-competitive inhibitors have also been reviewed recently , and these will only be discussed in so far as they give insights into functionally significant conformations.
2. Relating structure and function
2.1. The inactive monomeric CDK fold
CDKs vary in the lengths of N- and C-terminal sequences that bookend the conserved, central protein kinase domain  (figure 1). Overall, the structures of cyclin-free CDK1 (, PDB 4YC6), CDK2 (, PDB 1HCK), CDK6 (, e.g. PDB 5L2S), CDK7 (, PDB 1UA2) and CDK16 (, PDB 5G6 V) superimpose very well. For example, monomeric CDK2 and CDK7 overlay with an r.m.s.d. (root-mean-square deviation) of 1.49 Å over 262 equivalent Cα atoms. They share conserved structural features that ensure they are catalytically inactive (figure 2a). The start of the activation loop (defined as the sequence between the conserved DFG and APE motifs, residues 145–172 in CDK2) adopts a short α-helical conformation (αL12) that blocks the C-helix from swinging in to reshape the back of the active-site cleft. A characteristic of the glycine-rich (residues 12–16 in CDK2 that encodes the conserved GXGXXG motif) and activation loops is their relative mobility. As a result, differences between cyclin-free CDK structures are most evident around the active site (figure 2b,c). Accompanying these changes are more subtle differences in the relative dispositions of the N- and C-terminal lobes that lead to other conserved residues within the catalytic sites adopting positions that are incompatible with catalysis (figure 2b).
The classical model of CDK activation exemplified by CDK2–cyclin A is not applicable to the CDK5-related sub-branch of the CDK family of which CDK16 is a member . There are several emerging unusual features of CDK16 activation that would benefit from structural characterization. A CDK16 feature that it shares with CDKs 14, 15, 17 and 18 is an extended N-terminal regulatory region before the start of the kinase domain. This sequence is important for CDK16 association with its cognate cyclin, cyclin Y or cyclin Y-like 1 [3,41–43]. In addition, stable association of cyclin Y with either CDK14  or CDK16  requires cyclin Y phosphorylation and binding to 14-3-3, suggesting that a classical bidentate 14-3-3–ligand interaction  may help to organize cyclin Y to bind to its cognate CDK partner.
2.2. CDK2–cyclin A activation
CDK2 partners cyclin E during late G1 and is subsequently bound to cyclin A during S-phase for DNA replication . A series of structures of CDK2 bound to cyclin A provided snapshots of the structural changes that accompany cyclin binding and phosphorylation of the CDK2 activation loop [39,47,48] (figure 3a). Subsequent studies that have interrogated the kinetics of CDK2 activation in a cellular context have demonstrated that CDK-activating kinase (CAK, a complex of CDK7 and cyclin H in humans) is active against CDK2 (i.e. through phosphorylation of CDK2 T160), which is then proposed to bind to cyclin A . This result suggests a model in which flexibility around T160 is required for CDK2 to be recognized by CAK and that the adoption of an ordered activation loop conformation accompanies anchoring of the phospho-threonine residue promoted by cyclin binding.
The flexibility of the CDK fold has also been captured in ATP-competitive inhibitor-bound structures where inhibitor binding helps to stabilize alternative energetically less favourable conformations. At the start of the activation loop, the conserved DFG motif can adopt either an active ‘DFG-in’ conformation (figure 3), or an inactive ‘DFG-out’ conformation in which the phenylalanine side chain points into the active-site cleft and is removed from its position in the ‘regulatory spine’ of residues that characterizes the active protein kinase fold . This latter conformation has been exploited for the design of several tyrosine kinase-specific inhibitors [54,55]. Though the majority of CDK ATP-competitive inhibitor structures determined to date have a ‘DFG-in’ conformation , inhibitor binding to monomeric CDK2 (, PDB 5A14) and monomeric CDK16 (, PDB 5G6 V) and to cyclin-bound CDK8 (PDB 3RGF) can stabilize the CDK fold into a ‘DFG-out’ conformation. Thus, the binding of ATP-competitive inhibitors interrogated by the determination of multiple ‘snapshots’ of protein kinase structures highlights the inherent flexibility of the CDK fold and its ability to adopt multiple conformations [31,55].
2.3. Extending the activation model to other cyclin partners of CDK1 and CDK2
CDK1 is the closest member of the CDK family to CDK2 and for which structures of the cyclin-free and authentic cyclin-bound forms can also be compared (figure 3b; , PDB codes 4YC6 and 4YC3). It is the only essential CDK and, activated by its partners cyclins A and B, it executes progression through mitosis. Overall, the mechanism of CDK1 activation is conserved with CDK2. However, an opening of the interface coupled with a twist between the two proteins relative to CDK2–cyclin A results in a re-orientation of the C-helix and fewer interactions between the cyclin B and CDK1 C-terminal lobes. Overall, the interfacial surface is 30% smaller in CDK1–cyclin B compared with CDK2–cyclin A. Crystallographic electron density maps of unphosphorylated CDK1 suggest that it has a more flexible activation segment than does the comparable state of CDK2.
A comparative analysis of the sequence loci that mediate the CDK1– and CDK2–cyclin interfaces reveals the conserved sequence features that may explain CDK1 and CDK2 cyclin selectivity . CDK2 is partnered by cyclin E during late G1 phase and then subsequently by cyclin A . Under circumstances where CDK1 expression is knocked down, it can also partner cyclin B . A comparison of the structures of phosphorylated CDK2 bound to cyclin A (, PDB 1JST), cyclin B , PDB 2JGZ) and cyclin E (, PDB 1W98) revealed the conserved nature of the CDK2 response to cyclin binding . Cyclins A and B conserve three large aromatic residues at the CDK–cyclin interface (Y170, Y177 and Y258 in cyclin B), whereas in cyclin E the residues at these positions have smaller side chains (N112, I119 and L202). Given the smaller CDK1–cyclin interface compared with CDK2–cyclin A, the structures would predict that CDK1 would bind preferentially to cyclins B and A, but that these smaller side chains would have less impact on CDK–cyclin affinity in the context of the larger CDK2–cyclin interface.
A comparison of the CDK1–cyclin B and CDK2–cyclin A/B/E structures also highlights the potential for these closely related CDKs to be differentially regulated by reversible phosphorylation. The antagonistic activities of Wee1/Myt1 kinases and Cdc25 phosphatases regulate the phosphorylation status of the CDK glycine-rich loop (defined by the GXGXXG motif, residues 11–16 in CDK2). The structure of CDK2–cyclin A phosphorylated on Y15 illustrates how phosphorylation promotes a glycine loop structure that antagonizes both peptide substrate binding and the ATP conformation required for catalysis . The flexibility of the glycine-rich loop is compatible with a model in which the phosphorylated Y15 side chain is solvent exposed and accessible to both kinases and phosphatases. CDK1 is also regulated by active-site phosphorylation, and the conserved nature of the structure in this region suggests that the mechanism of inhibition is also conserved.
However, unlike the glycine-rich loop, the flexibilities of the phosphorylated CDK1 and CDK2 activation loops differ. Though the structure of a T161-phosphorylated CDK1–cyclin B complex is yet to be determined, this complex is susceptible to phosphatase treatment, suggesting that the phosphorylated CDK1 activation segment remains flexible . By contrast, phosphorylated CDK2 T160 is embedded within a network of ionic interactions (figure 3) that orders the CDK2 activation segment within this region and decreases T160 solvent accessibility. Taking Y15 as the model, this difference could ensure that the activity of CDK1, more so than that of CDK2, remains subject to the ongoing antagonistic activities of CAK and phosphatases. In particular, it would offer an opportunity for CDK1 to be subject to rapid enzyme-mediated inactivation even in the presence of high concentrations of cyclin B and might offer a regulatory opportunity to distinguish CDK1 and CDK2 activities.
Ringo/Spy proteins also activate CDK1 and CDK2 and represent a divergent branch of the cyclin family, identified through their ability to induce meiotic maturation in Xenopus oocytes [60,61], an activity conserved in humans . Ringo A/Spy1 is required for localizing CDK2 to telomeres, and its absence results in defects in chromosome tethering to the nuclear envelope [63,64]. Several studies have implicated Spy1 in glioma, suggesting that it may also have functions in mitosis in selected cell types . Ringo A knockout mice show similar defects to CDK2 knockout mice during spermatogenesis , suggesting that the essential function of CDK2 during meiosis might be mediated, in part, by its association with Ringo A. Spy1 (Ringo A) encodes only a single CBF embedded within a longer sequence and activates CDK2 through a mechanism that does not require activation loop phosphorylation (figure 3c; , PDB 5UQ2). Immediately after the DFG motif, CDK2 R157 and T158 anchor the activation loop through electrostatic interactions with Spy1 D97 and E135, respectively. CDK2 R50 and R150 that coordinate the phosphorylated CDK2 T160 side chain in the CDK2–cyclin A structure interact with Spy1 D136, so that its carboxylate moiety effectively mimics a number of interactions made by the phosphoT160 phosphate group. These alternative interface interactions create a CDK2 activation loop conformation most reminiscent of that seen when it is bound to cyclin A (figure 3d). The resulting complex has measurable kinase activity but is less active than phosphorylated CDK2–cyclin A .
3. CDK substrate recognition
The structure of CDK2–cyclin A bound to a non-hydrolysable ATP analogue and an optimal substrate peptide (HHASPRK) revealed how the activation segment is modelled to recognize a proline residue at the P + 1 position and a positively charged residue at P + 3 (where P is the phosphate-accepting residue) (, PDB 2CCI; figure 3d). Structural studies support a dissociative mechanism through a metaphosphate intermediate in which the attacking group (serine or threonine hydroxyl) from the peptide substrate comes in opposite to the leaving group (phosphate ester oxygen of the γ-phosphate group of ATP), leading to inversion of configuration at the phosphorus (PDB codes: 3QHR and 3QHW , and 1GY3 ). Apart from this motif, the only other significant sequence feature shared by many cell cycle CDK substrates is the RXL motif, first identified by comparative sequence analysis of multiple CDK substrates and inhibitors . This sequence binds to a site on the cyclin N-CBF that is conserved between cyclins A, B, D and E, and was first structurally characterized following the determination of the structure of a CDK2–cyclin A–p27KIP1 complex (PDB 1JSU, ).
A feature of the cyclin B-bound CDK1 is the retention of flexibility within the activation loop upon T161 phosphorylation  (figure 3b). Using a series of model peptide substrates, a comparative activity study suggested that for CDK1, this enhanced flexibility translates into a more relaxed substrate preference around the site of phospho-transfer . In the presence of an RXL motif, CDK1 will phosphorylate motifs that contain either a proline residue at P + 2 or a positively charged residue at P + 3. CDK1 is characterized by its promiscuous ability to phosphorylate a wide variety of substrates at multiple sites, many of which are ‘non-canonical’ [116,118–120]. The structure of CDK1 suggests a mechanism by which activation loop flexibility, embedded in an inherently, more flexible CDK1 fold allows CDK1 to accommodate a more diverse substrate set than its nearest relative CDK2. These plastic properties may also contribute to its ability to partner non-cognate cyclins in the absence of other CDKs to drive the cell cycle [34,121].
The structures of CDK4 bound to cyclin D1 and cyclin D3 support a model in which a catalytically competent active-site configuration must occur transiently when CDK4–cyclin D forms a Michaelis complex with ATP and protein substrates (figure 4b,c). Purified CDK4–cyclin D3 requires the presence of an RXL motif within the peptide substrate for activity, suggesting that substrate engagement through the cyclin recruitment site promotes both productive substrate engagement and kinase remodelling. Such a substrate-assisted catalysis model would be supported by kinetic studies in which CDK4 has been shown to follow an ordered sequential mechanism in which ATP binds first and the phospho-peptide product leaves last . CDK4/6–cyclin D complexes monophosphorylate pRB at multiple sites and further hyperphosphorylation is mediated by CDK2–cyclin E . Although it is not clear what function monophosphorylation performs, taken together, these observations suggest that CDK4 activity is more tightly regulated by substrate scaffolding than CDK1 and CDK2. Whether the model extends to CDK6 awaits the determination of the structure of CDK6 bound to an authentic D-type cyclin.
The RXL-binding cyclin recruitment site was the first to highlight the use of substrate docking sites to enhance CDK activity towards particular substrates [124–126]. Permutations on this sequence can be accommodated with differing affinities by cyclins to refine substrate recognition [58,127,128]. Compatible with a docking model, crystallographic attempts to determine a substrate path between the RXL and SPXK motifs for the binding of a model substrate to CDK2–cyclin A failed to resolve electron density for residues beyond the consensus sequences .
The ability of Cks1 to enhance the phosphorylation of a subset of CDK1 substrates was first recognized in Xenopus oocytes  and refined by further studies in Saccharomyces cerevisiae . Cks1 binds to the CDK1 C-terminal lobe (figure 6c) and contains a phospho-threonine docking site that can recognize phosphorylated CDK1 substrates and promote their further hyperphosphorylation by CDK1 . The order and pattern of target residue phosphorylation in multi-site phosphorylated substrates appears to be fine-tuned by the identity of the cyclin and the presence of Cks1 [131,133,134].
CDKs 7, 9, 12 and 13 phosphorylate the RNA polymerase CTD. The sequence of the CTD is unusual, being composed of 52 heptad repeats in humans, with the consensus sequence Y-S-P-T-S-P-S. Extracted from cells, CTD residues S2 and S5 are the most abundantly phosphorylated serine residues, while S7 is phosphorylated to a lesser extent [109,110]. The extent of phosphorylation within cells was found to be much less than expected, suggesting that multiple phosphorylation events within a single repeat or singly within adjacent repeats must be infrequent. Various studies have, together, suggested that the transcriptional CDKs have preferences for particular sites. For example, CDK7 has been shown to predominantly phosphorylate S5 and S7, CDK9 to have activity towards all three serines, and CDK12 and CDK13 to predominantly phosphorylate S2 . Functionally significant interplay between phosphorylation sites has been shown for CDK9 where, using model three hepta-repeat substrates, S7 phosphorylation was found to prime subsequent CDK9-mediated phosphorylation. In this study, pre-phosphorylation of S2 or S5 blocked subsequent CDK9 activity and CDK9 preferentially phosphorylated S5 . Unfortunately, there was no electron density to support binding of an S2 phosphorylated 13-mer substrate peptide following attempts to co-crystallize it with CDK13 . To date, there is no detailed structural information to understand the molecular determinants that distinguish the activities of the CTD kinases towards their shared substrate and to what extent the complex local molecular environment impacts substrate selection.
Other CDK substrate docking sites have been identified but as yet structural information is lacking. Analysis of a set of S. cerevisiae Cln2 mutants has identified a surface shared with Ccn1 and Cln1 cyclin subtypes but not with Cln3 that recognizes a consensus substrate ‘LP motif’ that is enriched in leucine and proline residues . Modelling the Cln2 structure on cyclin A reveals the docking site to be adjacent but non-overlapping with the RXL-binding site on the surface of the N-CBF. It is likely that ordered progression through the cell cycle results both from different CDK–cyclin pairings having different substrate selectivity and from the fact that the different CDK–cyclin pairings are expressed at different points in the cell cycle  (reviewed in ).
4. Regulatory protein interactions
4.1. Cell cycle CDK–cyclins: regulatory interactions determining activity
A number of cyclin-encoded protein-binding sites or short peptide motifs have been structurally characterized. A well-characterized example is the recycling of the cyclin RXL recruitment site that is exploited to either enhance or inhibit CDK activity. Alternatively, short motifs encoded within the cyclin sequence can be used both to dock cyclins to substrates to enhance CDK activity and alternatively to localize them to CDK regulators frequently resulting in a loss of CDK activity.
Members of the p27KIP1/p21CIP1 cyclin-dependent kinase inhibitor (CKI) family share an RXL motif with RXL-containing substrates and compete with them for CDK–cyclin association. The structure of a CDK2–cyclin A–p27KIP1 complex (PDB 1JSU, ) revealed the extended path of the N-terminal sequence of the intrinsically disordered p27KIP1 protein over the upper surface of the cyclin N-CBF (figure 6a). p27KIP1 then proceeds to disengage the edge β2-strand from the CDK2 N-terminal lobe and occupy the ATP-binding site, mimicking the interactions made by the adenine ring of ATP. p27KIP1 also acts as an assembly factor during G1 to assist the formation of active CDK4/6–cyclin D complexes, a role that also sequesters p21CIP1/p27KIP1 CKIs to promote G1 progression [27,139]. The retention of CDK activity in the presence of bound p27KIP1 is linked to the phosphorylation status of p27KIP1 Y88. Phosphorylation by tyrosine kinases (e.g. Src or Abl kinases) can generate CDK4/6–cyclin D–p27KIP1 [140–142] or CDK2–cyclin A–p27KIP1  complexes that are catalytically active. The differences in kinetics and affinity of p27KIP1 and p21CIP1 binding to CDK2–cyclin A and to CDK4–cyclin D complexes may reflect an option for an alternative binding mode to CDK4 [144–146]. Exploiting NMR methods, p27KIP1 Y88 phosphorylation promotes the removal of the 310 helix that occludes the CDK2 active site . The structural basis of how phosphorylated p27KIP1 binds to CDK4/6–cyclin D to aid assembly of an active complex is yet to be elucidated by a co-complex structure.
The INK (inhibitors of CDK) family of CKIs selectively inhibits CDK4 or CDK6 and, through an allosteric mechanism, disfavours CDK–cyclin binding . Their tandem ankyrin repeat structures exemplified by CDK6–p19INK4d (, PDB 1BLX; , PDB 1BI8) and CDK6–p16INK4a (, PDB 1BI7) bind in the vicinity of the CDK hinge on the interface opposite to the surface remodelled upon cyclin association (figure 6b). INK4 binding to CDK6 distorts the N-terminal kinase lobe relative to the C-lobe by approximately 15°, thus misaligning the key catalytic residues. The structures of individual INKs have also been determined by X-ray crystallography (p18INK4c, , PDB 1IHB) and (p19INK4d, , PDB 1BD8) or solution NMR (p15INK4b, , PDB 1D9S), (p16INK4a, , PDB 1DC2; p18INK4c, , PDB 1BU9; and p19INK4d, , PDB 1AP7).
The cell cycle CDKs are further distinguished by the CDK surfaces they exploit to regulate activity. For example, no protein equivalent to the INKs has been reported to bind to the CDK1/2 hinge. Similarly, there is no known protein that binds to CDK4 and CDK6 in a manner equivalent to the binding of Cks1 or Cks2 to CDK1 (, PDB 4YC6; figure 6c) or CDK2 (, PDB 1BUH). The CDK2 C-terminal lobe also recognizes kinase-associated phosphatase (KAP) that can dephosphorylate T160-phosphorylated CDK2 (, PDB 1FQ1; figure 6d).
In addition to helping to select mitotic substrate phosphorylation sites (see above), Cks1 collaborates with Skp2 to form the p27KIP1 phosphoT187-binding site within the SCFSkp2 (Skp1–cullin–F-box) E3 ubiquitin ligase complex (, PDB 2AST). This example is the first to show an F-box protein requirement for an accessory protein for substrate recognition [157,158]. Modelling studies using structures of sub-complexes show that a CDK2–cyclin A–p27KIP1–Cks1–Skp1–Skp2 complex can be built , but whether any subtle rearrangements occur will require determination of the structure of the CDK2–cyclin A–pT187p27KIP1–SCFSkp2 complex.
The LXCXE motif located towards the N-terminus of the D-type cyclins is highly conserved and represents an interesting example of a short cyclin-encoded motif that assists in substrate recruitment. D-type cyclins share this sequence with other cellular and viral proteins that bind to pRB . In the CDK4–cyclin D1 structure, the motif is sequestered in the channel between the C-terminal CDK and cyclin lobes (figure 4b). However, the quality of the electron density map shows that it is flexible, suggesting it could disengage and remodel to bind to pRB. The structure of a complex of the pRB pocket domain and an LXCXE-containing peptide derived from the human papilloma virus E7 protein illustrates the interaction (, PDB code 1GUX). It is not known whether pRB and cyclin D engagement of LXCXE and RXL motifs, respectively, is synergistic or antagonistic for promoting pRB phosphorylation by CDK4 or CDK6, but it may be hypothesized to contribute to the mechanism that restricts CDK4/6 activity. Mutation of the LXCXE motif disrupts cyclin D1 activity in some cell line contexts where cyclin D expression has been reduced , but its mutation in a cyclin D1 ‘knock-in’ mouse study did not reveal any significant differences to the authentic cyclin D1 sequence .