Cyclin-dependent kinases (CDKs) are serine/threonine kinases whose activity depends on a regulatory subunit - a cyclin. Based on the sequence of the kinase domain, CDKs belong to the CMGC group of kinases (named for the initials of some members), along with mitogen-activated protein kinases (MAPKs), glycogen synthase kinase-3 beta (Gsk3β), members of the dual-specificity tyrosine-regulated kinase (DYRK) family and CDK-like kinases [1]. In related kinases such as MAPKs, substrate specificity is conferred by docking sites separated from the catalytic site, whereas CDKs are characterized by dependency on separate protein subunits that provide additional sequences required for enzymatic activity. To aid nomenclature and analysis of CDKs, proteins belonging to this family have been recently renamed as Cdk1 through to Cdk20 [2].

CDKs were first discovered by genetic and biochemical studies in model organisms such as yeasts and frogs (reviewed in [3]). This work established the importance of CDKs in promoting transitions through the cell cycle. In addition, these studies showed that the catalytic subunit, the CDK, must associate with a regulatory subunit, the cyclin, whose protein levels are subject to regulation during the cell cycle (this oscillation lent these regulators their cyclin name). Since these pioneer studies conducted in the 1980s, the importance of CDKs acting as a major eukaryotic protein kinase family involved in the integration of extracellular and intracellular signals to modulate gene transcription and cell division has been clearly established [3–6].

Despite their function in eukaryotic cell division and transcription, CDKs have undergone an extraordinary degree of evolutionary divergence and specialization. Six different CDKs are present in budding yeast (Figure 1). These CDKs can be grouped as, first, CDKs that bind multiple cyclins and can regulate the cell cycle and, second, CDKs that are activated by a single cyclin and are involved in the regulation of transcription. In the budding yeast Saccharomyces cerevisiae, the first group contains Cdc28 and Pho85, each binding nine or ten different cyclins, respectively. This promiscuity forms the basis for their dynamic regulation and their ability to phosphorylate multiple substrates, thus regulating the cell-division cycle in response to different cellular cues. The second group comprises four CDKs - Kin28, Srb10, Bur1 and Ctk1 - each activated by a single specific cyclin (Figure 1). These cyclins are usually not regulated in a cell-cycle-dependent manner, and the members of this second group of CDKs are involved in the control of gene transcription.

The number of CDKs increased during evolution and was marked by a greater expansion of the cell-cycle-related group. Fungi contain 6 to 8 CDKs and 9 to 15 cyclins, whereas flies and echinodermata contain 11 CDKs and 14 cyclins, and human cells have 20 CDKs and 29 cyclins (Box 1) [7]. Evolutionary studies suggest that CDKs fall into eight subfamilies represented by Cdk1, Cdk4 and Cdk5 (from the yeast cell-cycle-related CDKs), and Cdk7, Cdk8, Cdk9, Cdk11 and Cdk20 (functioning as transcriptional CDKs) [7, 8] (Figure 2). Like its yeast ortholog, Cdk1 is the only CDK essential for the cell cycle in mammals [9], whereas both Cdk2 and Cdk3 are dispensable [3, 10]. Although Pho85 is not essential in yeast, this kinase is required for viability in some stress conditions, such as growth after starvation. Pho85 displays multiple cell-cycle-related functions as well as regulation of gene expression, metabolism, morphogenesis, cell polarity and aging; it functions as an integrator of signals such as nutrient availability, DNA damage or other types of stress [11]. Sequencing and functional studies suggest that the mammalian homolog of Pho85 is Cdk5, although these kinases cluster with multiple mammalian kinases of the Cdk5 subfamily, namely Cdk14 to Cdk18. Pho85 can interact with up to 10 cyclins of the Pcl1/Pcl2 or Pho80 groups, whereas mammalian Cdk5 is activated by non-cyclin proteins, including Cdk5R1 (p35) and Cdk5R2 (p39). Interestingly, other members of the Cdk5 subfamily, such as Cdk14 or Cdk16, are activated by cyclin Y, which is a cyclin closely related to yeast Pcl1/Pcl2 proteins [12, 13]. The Cdk4 subfamily is unique as it is only found in eumetazoans, and the members of this family diverge equally from the Cdk1 or Cdk5 subfamilies (Figures 1 and 2) [7]. Other cell-cycle-related subfamilies, such as the Cdk1-related B-type CDKs, are plant specific and are not found in animals or fungi [14].

Transcriptional CDKs are more conserved, both in sequence and function (Figure 1). Yeast Kin28 and human Cdk7 are subunits of transcription factor TFIIH, which is involved in transcription initiation by phosphorylating the Ser5 residue of the RNA polymerase II (RNAPII) C-terminal domain (CTD) at gene promoters. Cdk7 is also able to phosphorylate and activate other CDKs, thus acting as a CDK-activating kinase (CAK; Box 2). Kin28 does not have this activity, which is mediated in yeast by a different kinase unrelated to CDKs, Cak1 [8]. The yeast protein Srb10 is orthologous to human Cdk8 and Cdk19 and is the enzymatic component of the Mediator complex involved in the regulation of RNAPII during transcription [15]. Cdk9 is the yeast Bur1 ortholog, whereas the function of yeast Ctk1 in the phosphorylation of the RNAPII CTD is performed by Cdk12 in Drosophila and in human cells [16]. The evolutionary relationship of the Cdk11 and Cdk20 subfamilies to the yeast CDKs is not clear, although these proteins are well conserved [7]. Unlike cyclins for cell-cycle-related kinases, the cyclin subunits of transcriptional CDKs do not show significant oscillations in protein levels during the cell cycle, and these transcriptional CDKs are therefore regulated by protein-protein interactions or other mechanisms. Transcription-related kinases possibly originated after cell-cycle-related CDKs and became more diverse as the complexity of transcription increased [17].

Like other CMGC kinases, CDKs are proline-directed serine/threonine-protein kinases with some preference for the S/T-P-X-K/R sequence as a consequence of the presence of a hydrophobic pocket near the catalytic site that accommodates the proline (position +1). However, the requirement for the basic residue in the +3 position is not maintained in Cdk4 or transcriptional CDKs, which display a less-stringent S/T-P-X consensus. Some other family members such as Cdk7 or Cdk9 are not necessarily proline directed and can also phosphorylate residues in the absence of the +1 proline [18].

The CDKs range in size from approximately 250 amino acid residues, just encompassing the catalytic serine/threonine kinase domain, to proteins of more than 1,500 residues, with amino- and/or carboxy-terminal extensions of variable lengths (Figure 2). Like all kinases, CDKs have a two-lobed structure. The amino-terminal lobe contains beta-sheets, whereas the carboxy-terminal lobe is rich in α-helices, and the active site is sandwiched in-between. The N-lobe contains a glycine-rich inhibitory element (G-loop) and a unique major helix - the C-helix (containing the PSTAIRE sequence in Cdk1). The C-lobe contains the activation segment, which spans from the DFG motif (D145 in Cdk2; EMBL:AK291941) to the APE motif (E172 in Cdk2) and includes the phosphorylation-sensitive (T160 in Cdk2) residue in the so-called T-loop (Figure 3). In the cyclin-free monomeric form the CDK catalytic cleft is closed by the T-loop, preventing enzymatic activity. In addition, the activation segment in the C-lobe - a platform for binding of the phospho-acceptor Ser/Thr region of substrates - is partially disordered.

Upon binding of the cyclin to Cdk2, the CDK C-helix packs against one specific helix in the cyclin partner through a surface characterized by extensive hydrophobic interactions. Association of cyclins to the C-helix promotes a rotation in the axis of this segment, generating new interactions that are part of the active ATP-binding site. In addition, cyclins take the C-lobe activation segment out of the catalytic site so that the threonine becomes accessible for activating phosphorylation by CAK (Figure 3). This phospho-threonine acts as a rigidifying hub, stabilizing the activated form of the kinase heterodimer [18, 19]. The extent of the CDK-cyclin interface varies in the structure of Cdk4, Cdk9 or yeast Pho85 [18, 20, 21]. For instance, Cdk2 and cyclin A contact each other at both the N- and C-lobes, whereas the contacts between Cdk4 and cyclin D are limited to the N-lobe, and, unlike Cdk2, the cyclin does not impose an active conformation on the kinase as the Cdk4 ATP-binding site is still inaccessible to its substrates, even in the presence of the cyclin [20, 21]. How Cdk4 becomes active is not completely clear, although the binding of the substrate is thought to induce the activation segment to open and to fit to the phospho-acceptor site. Some CDKs, such as Cdk5 or its yeast ortholog Pho85, do not require phosphorylation in the activation segment for activity, and these kinase can adopt the correct conformation through other interactions [18].

In addition to the consensus kinase domain, a few CDKs contain additional domains with functional relevance. Cdk16, Cdk17 and Cdk18 (containing a PCTAIRE sequence in the C-helix) are characterized by a conserved catalytic domain flanked by amino- and carboxy-terminal extensions involved in cyclin binding. Phosphorylation of the Cdk16 amino-terminal domain blocks binding to cyclin Y, providing a novel mechanism for regulation of these complexes [22]. In Cdk12 and Cdk13 (characterized by a PITAIRE motif), the kinase domain is localized in the center, and additional Arg/Ser-rich motifs in the amino terminus serve as docking sites for the assembly of splicing factors and regulators of splicing (Figure 2). These two kinases also contain proline-rich motifs, mostly concentrated in their carboxy-terminal region, that serve as binding sites for Src-homology 3 (SH3), WW or profilin-domain-containing proteins [16].

The glycine-rich region (G-loop) in the N-lobe is an additional regulatory region as it contains residues (Thr14 and Tyr15 in Cdk2; Figure 3) whose phosphorylation inhibits kinase activity. Phosphorylation of Thr14 and/or Tyr15 residues by Wee1 and Myt1 kinases inhibits several family members, preventing cell-cycle progression, for instance, in response to DNA damage. Elimination of these phosphates by phosphatases of the Cdc25 family is then required for activation of CDKs and cell-cycle progression [3, 23]. Inhibitory phosphorylation at Thr14 and Tyr15 does not result in major changes in the CDK structure, but does inhibit the CDK activity by reducing the affinity of the CDK for its substrates. However, phosphorylation at Tyr15 seems to be activating in the case of Cdk5, perhaps by improving substrate recognition [18]. These residues are not present in Cdk7, in agreement with the general belief that this kinase is constitutively active and regulated at different levels.

Cell-cycle-related CDKs can also be negatively regulated by binding to small proteins of the INK4 or Cip/Kip families of inhibitors [19, 24]. INK4 proteins (p16 ^INK4a , p15 ^INK4b , p18 ^INK4c and p19 ^INK4d ) are specific for the Cdk4 subfamily and interact with the monomeric CDKs. They function by distorting the cyclin interface and the ATP-binding pocket, thus preventing activation of Cdk4 and Cdk6 by D-type cyclins or by CAK [24]. Members of the Cip/Kip family of inhibitors (p21 ^Cip1 , p27 ^Kip1 and p57 ^Kip2 ) contact both the CDK and cyclin subunits and are able to inhibit CDK-cyclin heterodimers, giving additional levels of regulation once these complexes have already formed [19].

It is abundantly clear that the CDK family is central to multiple signaling pathways controlling transcription and cell-cycle progression. CDKs probably originated as a system to modulate cell-cycle-promoting activity in response to various cellular scenarios. Over the course of evolution, both CDK and cyclin gene families have independently undergone a significant number of functional specializations [7]. Many of the interactions between specific mammalian CDKs and cyclins have been reported in vitro. However, the biochemical promiscuity in CDK-cyclin interactions makes it difficult to evaluate properly the in vivo physiological relevance of specific CDK-cyclin complexes. For instance, Cdk1 is thought to be activated mainly by A- and B-type cyclins but can also bind to, and be activated by, D- or E-type cyclins in the absence of Cdk4/Cdk6 or Cdk2, respectively [9, 27, 56]. Cdk5 can also bind to D-type cyclins, although to what extent these complexes are active or relevant in vivo is not clear. The situation is even more complex for the lesser-known family members for which there are no current in vivo data [2].

Although the comparison of the yeast CDKs has promoted the convenient division between transcriptional and cell-cycle activities, the multiple interactions between these two activities in higher eukaryotes makes it difficult to maintain this simple classification. First, transcription and cell-cycle progression cannot be opposed as these processes function at different layers in cell biology. Arguably, transcription is a major regulatory pathway required for cell-cycle entry. Major cell-cycle-related kinases such as Cdk4 and Cdk6 mostly function by phosphorylating transcription regulators such as Rb or Smads [3, 25], and the archetypal cell-cycle kinase Cdk1 also phosphorylates multiple transcription factors and epigenetic modulators (Figure 4) [5]. By contrast, major ‘transcriptional’ CDKs such as Cdk7 or Cdk11 directly control cell-cycle progression, in some instances independently of transcription. Finally, a single CDK can have separate cell-cycle-related and transcriptional activities. As an example, Cdk6 has recently been characterized as a chromatin factor (Figure 4) that regulates transcription factors involved in angiogenesis or the NF-κB pathway [57, 58], a process independent of the classical Cdk4/6-cyclin-D-Rb pathway involved in cell-cycle regulation.

As a consequence of their importance in multiple processes, CDKs are frequently mutated or deregulated in disease. A classic example is the almost universal deregulation of the CDK-cyclin-Rb pathway in cell-cycle entry during malignant transformation [25]. Underlining the significance of CDKs, inhibitors of Cdk4 and Cdk6 received in 2013 the Food and Drug Administration ‘breakthrough therapy’ designation for treatment of patients with breast cancer [59]. Other members of the CDK family can also be considered as interesting targets for therapeutics in cancer or other diseases. Cdk5 displays multiple roles in neurodegenerative diseases [28] and in other tissues with relevance to diabetes, cardiovascular disease or cancer [29]. Cdk8 exhibits copy-number gains in colon cancers, and recently it has been characterized as a coactivator of the beta-catenin pathway in colon cancer cell proliferation [60, 61]. Cdk10 is a major determinant of resistance to endocrine therapy for breast cancer [62], and inhibition of Cdk12 confers sensitivity to inhibitors of poly (ADP-ribose) polymerases PARP1 and PARP2 [63]. Cdk14 confers motility advantages and metastatic potential in hepatocellular carcinoma motility and metastasis [64, 65]. Finally, as indicated above, cyclin Y kinases regulate the Wnt pathway [31], providing new therapeutic opportunities that are yet to be explored. Hence, it seems very likely that new targets within the CDK family will be explored in the near future for therapy of cancer or other diseases.

Cyclins are a large family of approximately 30 proteins varying in mass from 35 to 90 kDa. These proteins are structurally defined by the presence of the so-called cyclin box, a domain of approximately 100 amino acid residues that forms a stack of five α-helices. Many cyclins have two cyclin boxes, one amino-terminal box for binding to CDKs, and a carboxy-terminal box that is usually required for the proper folding of the cyclin molecule. The cyclin box is also present in other molecules such as the retinoblastoma protein (Rb), the transcription factor TFIIB and Cables (CDK5 and ABL1 enzyme substrate 1), which are unlikely to function as CDK activators. In general, cyclins show less sequence similarity than the CDKs. The cyclin family contains approximately 29 protein in humans, clustered in 16 subfamilies and three major groups: group I (cyclin B group: A-, B-, D-, E-, F-, G, J, I and O); group II (cyclin Y group - a partner of the Cdk5 subfamily); and group III (cyclin C group: C-, H-, K-, L- and T- - major partners of transcriptional CDKs) [7, 66]. Cyclin D and cyclin E clades (partners of Cdk1 and Cdk4 subfamilies) have undergone lineage-specific expansion and specialization in metazoa and plants [7].

The CAK complex (comprising Cdk7, cyclin H and Mat1) phosphorylates the T-loop of all CDKs tested, thus participating in their activation. Furthermore, this complex can be part of the transcription factor phosphorylating the CTD of RNAPII as well as multiple nuclear receptors such as retinoic acid or thyroid receptors, the estrogen receptor α or the vitamin D receptor co-activator Ets1 [33]. The CAK complex can also be found associated with an additional subunit of TFIIH - the DNA-dependent helicase Xpd - forming a complex known as CAK-XPD. This complex plays a role in the coordination and progression of mitosis, likely as a consequence of the redistribution of CAK within different cell compartments during the late nuclear-division steps [67].