|Home | About | Journals | Submit | Contact Us | Français|
In contrast to the active conformations of protein kinases, which are essentially the same for all kinases, inactive kinase conformations are structurally diverse. Some inactive conformations are, however, observed repeatedly in different kinases, perhaps reflecting an important role in catalysis. In this review, we analyze one of these recurring conformations, first identified in CDK and Src kinases, which turned out to be central to understanding of how kinase domain of the EGF receptor is activated. This mechanism, which involves the stabilization of the active conformation of an α helix, has features in common with mechanisms operative in several other kinases.
Protein kinases are crucial elements of the signaling pathways that control cellular function. Kinases with specificity for either serine/threonine or tyrosine share a highly conserved catalytic domain that adopts a conformation when active that is also highly conserved (Hanks et al., 1988; Hubbard and Till, 2000; Knighton et al., 1991; Manning et al., 2002). What differentiates one kinase from another is the diversity of input signals that impinge on the catalytic domain, and a rich variation in the mechanisms that convert inactive forms of the kinase to active ones. These differences have been the key to the ability to target specific kinases by small molecules, underlying their growing importance in cancer therapy.
In an insightful review Louise Johnson and colleagues paraphrased the opening line of Tolstoy's Anna Karenina as a metaphor for understanding kinase regulation: “All active kinases are alike but an inactive kinase is inactive after its own fashion” (Noble et al., 2004). Free from the constraints of catalyzing the phosphate transfer reaction, the inactive forms of kinases can adopt radically different conformations around the active site, each uniquely specialized for responding to input signals. That the inactive conformations could be targeted specifically by small molecules was first visualized for a MAP kinase (Wang et al., 1998) and was highlighted by the discovery that the cancer drug imatinib (Gleevec, Novartis) recognizes a distinctive inactive conformation of its targets Abl and c-Kit, and that this feature underlies its specificity (Mol et al., 2004; Schindler et al., 2000).
In the few years that have passed since the Johnson review the number of protein kinase structures that have been determined has exploded (Eswaran and Knapp, 2010). From this harvest of molecular detail a new realization has emerged: the inactive conformations of kinases may fall into a relatively small number of classes, within each of which certain key features of the inactivation mechanism are conserved. This is not, in retrospect, surprising. Because protein kinases are subject to the physical constraints of the same protein fold, there are perhaps only a limited number of ways in which the fold can be distorted away from the active structure. It may even be that various “inactive” structures represent the stabilization of conformations that are intermediates in as yet poorly understood aspects of catalytic mechanism, such as nucleotide release, and so are very broadly conserved because they have a fundamental role in the phosphate transfer reaction.
Despite the presence of some common features in classes of inactive structures, it is still the case that because the structure need not be catalytically competent, each individual inactive kinase conformation is different in detail from other structures. Compounds targeting inactive conformations therefore provide increased opportunity for specificity compared to those that target the active conformation. Most kinase-driven diseases, such as cancers, typically involve the inappropriate activation of a kinase and it might seem counterintuitive to target inactive conformations. But kinases are highly dynamic, and are constantly switching between different conformations, and this process is further stimulated by the action of phosphatases that undo the action of activating phosphorylation events. Inhibition of the kinase can therefore be achieved by trapping it either in an active conformation (e. g. dasatinib (Tokarski et al., 2006)) or an inactive one (exemplified by imatinib).
One drug that targets the inactive conformation of a kinase is lapatinib, which inhibits the epidermal growth factor (EGF) receptor and is in current clinical use for breast cancer (Spector et al., 2005). Indeed, it was the elucidation of the structure of lapatinib bound to the EGF receptor kinase domain, by scientists at GlaxoSmithKline, that led to the realization that the EGF receptor could adopt this particular inactive conformation (Wood et al., 2004). This conformation was first identified in cyclin dependent kinases (CDKs) (De Bondt et al., 1993) and the Src family of kinases (Sicheri et al., 1997; Xu et al., 1997). This finding set the stage for unraveling how the kinase domain of EGF receptor is activated, which turned out to be quite different from the way that other receptor tyrosine kinases are controlled (Jura et al., 2009a; Red Brewer et al., 2009; Zhang et al., 2006).
The structures and regulatory mechanisms of protein kinases have been reviewed in depth elsewhere (Hubbard and Till, 2000; Huse and Kuriyan, 2002; Kornev and Taylor, 2010; Lemmon and Schlessinger, 2010; Noble et al., 2004; Pearce et al., 2010). Here, we focus on two particular aspects of kinases, viewed through the lens of the catalytic domain of the EGF receptor. First, we highlight the connection between the inactive conformations of the EGF receptor and those of the CDKs and the Src kinases. We point out that this class of inactive conformations is very broadly distributed, and discuss why this might be so. We then turn to the mechanism by which the kinase domain of the EGF receptor is activated, and make a connection to a mechanism of allosteric control that is very broadly distributed in kinases, including members of the AGC family, such as c-AMP dependent protein kinase A (PKA) and protein kinase B (PKB/AKT).
There are ~500 kinases in the human genome, with specificity towards serine/threonine or tyrosine, (the histidine kinases, prevalent in bacteria, are excluded from our discussion) (Manning et al., 2002). Most kinases have multiple regulatory domains that link kinase activity to unique signaling inputs and outputs (Pawson and Kofler, 2009). In spite of this broad diversity, the phosphorylation reaction itself constitutes a highly conserved process, which is dependent on a set of structural features of the active site that are common to all kinases.
Much of what we know about the structure of the active state of protein kinases emerged from studies on c-AMP-dependent protein kinase (PKA), a serine/threonine kinase (Kornev and Taylor, 2010). The active site is located between two lobes of the kinase domain (the N- and C-terminal lobes; Figure 1A). In the case of PKA, a hinge motion changes the relative orientation of these two lobes from a more open state in the inactive conformation to a more closed state in the active conformation. The closure of the lobes results in the proper positioning of ATP and the substrate in the active site and is dependent on phosphorylation of a threonine residue (Thr 197; the residue numbering and secondary structure notation are from PKA, PDB code: 1ATP; this numbering will be used in the rest of the review, unless otherwise specified) in a centrally located activation segment or “loop” (Knighton et al., 1991; Nolen et al., 2004). The general features of the mechanism of PKA are modulated to yield the diverse set of regulatory mechanisms seen in other kinases (Huse and Kuriyan, 2002).
The key structural features required for catalysis were revealed by the analysis of the crystal structures of the active conformation of PKA and the insulin receptor tyrosine kinase bound to ATP (or ATP-analogs) and substrate peptide (or mimics) (Bossemeyer et al., 1993; Hubbard, 1997; Knighton et al., 1991; Zheng et al., 1993). These structures highlighted the importance of the activation loop in controlling the activation state of the kinase. The N-terminal region of the activation loop contains a conserved Asp-Phe-Gly (DFG) motif. The sidechain of the aspartate in the DFG motif points towards the phosphate groups of ATP and plays a critical role in coordinating a magnesium ion, which is required for ATP binding. The C-terminal region of the activation loop adopts an open conformation, serving as a platform for docking the substrate peptide (Figure 1A). In some kinases, such as the C-terminal Src kinase, Csk, substrate docking involves interactions elsewhere, and the C-terminal portion of the activation loop is disordered (Levinson et al., 2008).
Another important structural element of the kinase active site is helix αC, in the N-lobe of the kinase, within which a conserved glutamate residue (Glu 91) is located. In the active conformation helix αC packs closely against the rest of the N-lobe of the kinase, allowing the glutamate residue to form a salt bridge with a conserved lysine residue (Lys 72) in strand β3 that coordinates the α- and β-phosphate groups of the substrate ATP molecule (Figure 1A). This lysine residue is commonly mutated to generate inactive forms of kinase domains (Robinson et al., 1996). Two other important interactions in the active site involve the glycine-rich P-loop and the highly conserved HRD motif (YRD in PKA) located in the catalytic loop that directly precedes the activation loop. The glycine-rich P-loop is important for nucleotide binding in the active site by making interactions with the β and γ-phosphates of ATP. The HRD aspartate (Asp 166) serves as a catalytic base to accept the proton from the hydroxyl group of the substrate residue during the catalysis (Figure 1A).
Two highly conserved and functionally important intramolecular networks between the N-lobe and the C-lobe are correlated with the activity of protein kinases (Kornev et al., 2006; Kornev et al., 2008). These networks, or “spines”, involve hydrophobic residues that can be assembled or disassembled depending on the presence of ATP or the substrate peptide (Figure 1B). One of these, the “regulatory spine”, is involved primarily in substrate binding and its proper assembly depends on the conformation of the activation loop (Kornev et al., 2006). There are four residues in the regulatory spine, Leu 106 in strand β4, Leu 95 in the catalytically important helix αC, Phe 185 in the conserved DFG motif and Tyr 164 in the conserved HRD motif. Upon activation, the change in activation loop conformation, which in most kinases is responsive to phosphorylation of the activation loop, changes the orientation of helix αC and the HRD motif, and results in the assembly of the regulatory spine. Alterations in the flexibility of the regulatory spine have been suggested to underlie the mechanism by which some of the frequently detected mutations in cancer patients mediate resistance to treatment with kinase inhibitors (Azam et al., 2008).
The second spine, the “catalytic spine”, incorporates the adenine ring of ATP and establishes a connection between the N-lobe and the C-lobe upon nucleotide binding (Kornev et al., 2008). The residues in the catalytic spine are located in helix αF (Met 231 and Leu 227) and helix αD (Met 128 and Leu 172) in the C-lobe, and the residues in the N-lobe: Ile 174 and Leu 173 in the β7 strand, Val 57 in β2 strand and Ala 70 in β3 strand. Helix αF also is also connected to the regulatory spine by an aspartate residue (Asp 220), which is highly conserved (Kornev et al., 2008). Helix αF, which is the most buried helix in the structure, emerges as an essential structural element in kinases that integrates assembly of the two hydrophobic spines with kinase activation (Figure 1B).
The discovery of the Src tyrosine kinase (initially characterized as its oncogenic form in the Rous Sarcoma virus) demonstrated a link between aberrant kinase activation and cancer (Martin, 2004). Much attention was therefore focused on understanding why the loss of a tyrosine residue in the C-terminal tail of c-Src (Tyr 527, chicken c-Src numbering) results in constitutive activation of the kinase. The crystal structures of two Src family kinases (c-Src and Hck) in the inactive conformation and the structure of active Lck revealed how the SH2 domains of these proteins help keep the kinase domains in an autoinhibited state, by binding to Tyr 527 (Sicheri et al., 1997; Xu et al., 1997; Yamaguchi and Hendrickson, 1996) (Figure 2A).
It was quite a surprise to discover that the inactive conformation of the c-Src and Hck kinase domains actually resembles the inactive conformation of the serine/threonine kinase, cyclin-dependent kinase 2 (CDK2), which was the first inactive conformation to be defined (De Bondt et al., 1993). This similarity is particularly striking because their regulatory mechanisms are so different (Figure 2A). Src kinases are autoinhibited by the regulatory SH2 and SH3 domains, in the absence of activating ligands or phosphorylation. In the case of CDKs, the kinase is by default in the inactive state and activation is achieved by the binding of cyclin proteins that are synthesized only at specific times during the cell cycle (Figure 2A). Because this inactive conformation was first discovered in CDKs and the Src kinases, we shall refer to it as the “CDK/Src-like” inactive conformation (Figure 2B). This inactive conformation has since been observed in many other serine/threonine and tyrosine kinases, such as Abl (Levinson et al., 2006), ZAP70 (Deindl et al., 2007), Wnk (Min et al., 2004), NEK2 (Rellos et al., 2007) and c-Met (Wang et al., 2006).
The CDK/Src-like conformation has the N- and C-lobes closed down over each other relative to active conformations (Figure 2B). In this closed conformation helix αC is swung outward from the N-lobe. This orientation of helix αC pulls the conserved glutamate sidechain in this helix (Glu 91 in PKA, Glu 310 in c-Src) out of the active site and disrupts its interaction with the conserved lysine residue (Lys 72 in PKA, Lys 295 in c-Src) from strand β3 in the N-lobe, leading to inactivation of the kinase. The movement of helix αC also disrupts the regulatory spine by removing the conserved hydrophobic residue in helix αC (Met 314 in Src) away from the active site.
Another important feature of the CDK/Src-like inactive conformation is that the portion of the activation loop immediately following the DFG motif often forms a short (single or double turn) helix (Figure 2B). In c-Src and Hck this particular conformation of the activation loop was observed in inactive structures that were determined subsequent to the first structures (Schindler et al., 1999; Xu et al., 1999). This helix stabilizes the swung-out conformation of helix αC by packing directly against it, using two or three conserved hydrophobic residues. The sidechains of these residues insert between Lys 295 (c-Src), in strand β3, and Glu 310 in helix αC, blocking the formation of the catalytically essential Lys 295 - Glu 310 salt bridge, which switches instead to form a Lys 295 – Asp 404 salt bridge with the DFG-aspartate (Figure 2B). Mutation of the hydrophobic residues in the helix activates c-Src (Gonfloni et al., 2000).
In the first inactive crystal structures solved of c-Src and Hck, the activation loop is in a slightly different conformation than that described above (Sicheri et al., 1997; Xu et al., 1997). A crystal structure of the initiation factor 2α protein kinase GCN2 shows a similar conformation (Padyana et al., 2005). In this conformational variant the N-terminal portion of the activation loop does not form the single turn helix (Figure 2B). The glutamate in helix αC forms an ionic interaction with the conserved arginine residue in the His-Arg-Asp (HRD) catalytic loop, instead of the arginine in the N-terminal portion of the activation loop. This conformation may represent an intermediate in the transition pathway from the typical CDK/Src-like to the active conformations, as discussed below.
The prevalence of the CDK/Src-like inactive conformation among distantly related kinases indicates that it might play some specific role in the general mechanism of kinases. One intriguing idea is that CDK/Src-like inactive conformation might be coupled to the “DFG flip”, a conformational change in the DFG motif in which the aspartate and phenylalanine sidechains exchange positions due to a crankshaft like motion of the peptide backbone (Figure 3A). The DFG flip from the DFG-in (active) to the DFG-out (inactive) conformation results in disruption of the regulatory spine by removing the phenylalanine from the core of the spine. The catalytic spine is also disrupted due to the loss of ATP binding in the nucleotide binding pocket, which is now occupied by the flipped phenylalanine. The resulting removal of the aspartate from the active site of the kinase prevents coordination of the magnesium ion that is required for catalysis. In the DFG-out conformation helix αC maintains its inward orientation and the glutamatelysine salt bridge. Crystal structures of the Abl and c-Kit kinases in complex with imatinib show that imatinib binding requires this DFG-out inactive conformation (Mol et al., 2004; Nagar et al., 2002; Schindler et al., 2000).
Src kinases can also readily adopt the DFG-out conformation, as demonstrated by a class of compounds, denoted the DSA compounds, that bind to c-Src and Hck with high affinity and require the DFG motif to be flipped. The DSA compounds are based on the chemical scaffold of imatinib. In contrast to imatinib, which binds to c-Src with a significantly lower affinity than to Abl, DSA compounds are equipotent inhibitors of c-Src and Abl (Seeliger et al., 2009). This means that the selective inhibition of Abl over c-Src by imatinib is not due to an impeded DFG flip in c-Src. Instead, differences in the P-loop of Abl and c-Src appear to underlie the specificity of imatinib for Abl over c-Src.
Just as the Src kinases can adopt the DFG-out conformation, Abl can adopt the CDK/Src-like inactive conformation (Levinson et al., 2006). This observation emphasizes the fact that one kinase can access multiple inactive conformations. Factors that perturb the free energy landscape of a kinase, such as the binding of an allosteric regulator or a kinase inhibitor, can tip the energetic balance and shift the kinase from one conformation to another.
What might be the significance of DFG flipping for kinase catalysis? Long time scale molecular dynamics simulations of the Abl kinase have yielded insights into the mechanism of the DFG flips (Shan et al., 2009). If one considers just the peptide segment spanning the DFG motif, it turns out that the preferred conformation corresponds to the flipped one (DFG-out), with the aspartate out of the active site. This is because the backbone ϕ and values in the DFG-in conformation are in an entropically disfavored region of the Ramachandran diagram, whereas the DFG-out conformation is in a more favorable region. The DFG-in conformation resembles a coiled spring that is ready to flip to a more favorable DFG-out conformation. The catalytic rate of protein kinases appears to be limited by the rate of ADP release (Grant and Adams, 1996; Lew et al., 1997). The nucleotide-free DFG-out conformation, which is more flexible than the DFG-in conformation, might facilitate nucleotide release and the rebinding of ATP (Shan et al., 2009).
The DFG flip results in the polar DFG-aspartate entering the hydrophobic environment previously occupied by the DFG-phenylalanine, which carries a high energetic penalty. Protonation of the DFG-aspartate, driven by the increase of its pKa upon ATP hydrolysis and the release of ADP and a magnesium ion from the active site, might decrease the associated cost in free energy (Shan et al., 2009). Due to the conservation of the DFG motif across the kinase family tree, this protonation-dependent switch might represent a general mechanism that facilitates the release of ADP from the active site.
The molecular dynamics simulations of the Abl kinase also point to a potential role of the CDK/Src-like conformation as an intermediate in DFG flipping (Levinson et al., 2006; Shan et al., 2009) (Figure 3B). The rotation of the bulky hydrophobic phenylalanine into the active site of the kinase domain during the DFG flip requires large-scale motions in the active site. Molecular dynamics simulations showed significant hinge-opening motions between the N-lobe and the C-lobe, which were associated with the movement of helix αC towards the CDK/Src-like inactive conformation (Figure 3C). These observations suggest that adopting the CDK/Src-like inactive conformation would facilitate the DFG-flip by enabling the DFG-phenylalanine to move towards the position previously occupied by the DFG-aspartate. One interesting aspect of this analysis is that intermediate structures in the computational trajectories correlate with the crystal structures of different kinases (Figure 3D). That is, intermediate steps in the DFG flip can be reconstructed by linking the experimentally determined structures of different kinases.
The activity of most protein kinases is enhanced by phosphorylation of the activation loop. Protein kinases are so constructed that this reaction requires one kinase molecule to phosphorylate another, although there is one notable exception – the DYR kinases have been shown to phosphorylate their activation loops in cis while partially unfolded during biosynthesis on the ribosome (Lochhead et al., 2005). Activation loop phosphorylation provides a simple mechanism for how the activity of receptor tyrosine kinases is controlled. In receptor tyrosine kinases, the kinase domain is coupled through a transmembrane domain to the extracellular ligand-binding domain. In a prototypical mechanism for receptor tyrosine kinase activation, ligand binding controls the ability of one kinase in a dimer to phosphorylate the other (Hubbard and Miller, 2007). This phosphorylation event is required to stabilize the active state of the kinase that enables efficient phosphorylation of other tyrosine residues in the receptor, primarily on the C-terminal tails of receptor itself. These phosphorylated tyrosine sites create binding sites for the SH2 and PTB domain-containing effector molecules that couple the activated receptors to downstream signaling pathways.
The EGF receptor (also known as HER1, for human EGF receptor or ErbB1 after the erythroblastoma viral gene) and its three close relatives in humans: HER2/ErbB2, HER4/ErbB4 and the catalytic inactive HER3/ErbB3, are critical regulators of mitogenic responses in cells and can elicit a potent oncogenic signal when deregulated in human diseases (Yarden and Sliwkowski, 2001). In fact, the discovery of the EGF receptor and its relative HER2 coincided with the first demonstrations that receptor tyrosine kinases are causally linked to cellular transformation. Following the identification of the receptor for EGF (Carpenter et al., 1975; Ullrich et al., 1984), Downward and colleagues discovered that the vErbB oncogene, carried by the avian erythroblastosis virus, is similar in sequence to the intracellular portion of the EGF receptor (Downward et al., 1984). Similarly, after human HER2 was identified (Coussens et al., 1985; Yamamoto et al., 1986), the p185Neu oncogene in rat fibroblasts was shown to be the rat homolog of human HER2 (Bargmann et al., 1986a, b).
The EGF receptor family members are quite divergent in their C-terminal tails and activate different sets of effectors proteins. One key feature of the activation of members of this family is the formation of both homo- and heterodimers, some of which include the inactive HER3, depending on the bound ligands. Since the catalytically inactive HER3 cannot activate its partner by phosphorylation, the mechanism by which these receptors become phosphorylated in heterodimers with HER3 was puzzling.
Another surprising aspect of the EGF receptor is the lack of requirement for activation loop phosphorylation. Although the EGF receptor contains a conserved phosphorylation site in its activation loop, Tyr 845 (human EGF receptor numbering), which undergoes rapid phosphorylation upon ligand binding (Biscardi et al., 1999), mutation of the activation loop tyrosine does not interfere with receptor activation (Gotoh et al., 1992; Tice et al., 1999). These observations focused attention on the idea that a key step in the activation of the kinase domain of the EGF receptor must involve an alternative mechanism, in which phosphorylation on Tyr 845 does not play a role. The fact that the kinase domains of the EGF receptor family are located closer on an evolutionary tree to those of non-receptor tyrosine kinases, such as ACK1 and Janus kinases (JAKs) (Manning et al., 2002), also suggests that the mechanism of the EGF receptor activation might be distinct from that described for other receptor tyrosine kinases.
Because the EGF receptor does not require activation loop phosphorylation, it was thought that this receptor is perhaps always in the active conformation and that dimerization of the receptor by ligand binding serves simply to enable trans phosphorylation of the C-terminal tails of the receptor. The finding that the kinase domain of the EGF receptor adopts the CDK/Src-like inactive structure when bound to lapatinib suggested that the kinase domain might be autoinhibited in some way prior to ligand binding (Wood et al., 2004). Also suggestive of the relevance of the CDK/Src-like inactive structure to the EGF receptor function was the discovery of mutations in the activation loop of the EGF receptor in some cancer patients (Lynch et al., 2004; Paez et al., 2004). These mutations, such as Leu 834 to Arg, have been shown to activate the kinase, most likely by destabilization of the inactive CDK/Src-like conformation (Yun et al., 2007; Zhang et al., 2006). We had noted earlier in this review that the corresponding mutations in c-Src result in its activation (Gonfloni et al., 2000).
The realization as to how the EGF receptor is activated came from analysis of several crystal structures of its kinase domain in the active conformation (Zhang et al., 2006), determined originally at Genentech (Stamos et al., 2002). In all of these structures the kinase domains form an asymmetric, head to tail, dimer, in which the binding of one kinase domain (denoted as the activator kinase) stabilizes the active conformation of the second kinase domain (denoted the receiver kinase) (Figure 4A). The dimerization interface is largely hydrophobic and involves the bottom of the C-lobe of the activator kinase, which docks on the top of the N-lobe of the receiver kinase.
The principal interactions in the asymmetric dimer are between helix αH of the activator kinase and helix αC of the receiver. This interaction stabilizes the swung-in conformation of helix αC in the receiver kinase and the extended conformation of the activation loop. The formation of this asymmetric dimer is necessary for the activation of the EGF receptor and also underlies the activation of other members of the EGF receptor family in homo- and heterodimers (Jura et al., 2009b; Monsey et al., 2010; Qiu et al., 2008). When the asymmetric dimer interface is disrupted, either by point mutations in the activator interface (such as Val 924 Arg) or by binding of the negative feedback inhibitor of the EGF receptor Mig6 to the activator interface, the EGF receptor kinase domain adopts the CDK/Src-like inactive conformation in crystal structures (Zhang et al., 2006; Zhang et al., 2007).
The activation of the EGF receptor kinase domain by formation of the asymmetric dimer is reminiscent of the way cyclin-dependent kinases CDKs become activated by their allosteric regulators, the cyclins (De Bondt et al., 1993; Jeffrey et al., 1995) (Figure 4B). In the crystal structure of the CDK2/cyclinA complex, cyclin binds to the active conformation of CDK, engaging the N-lobe, activation loop and the C-lobe of CDK (Figure 4B) (Jeffrey et al., 1995). One of the major interactions between cyclin and CDK is the packing of helix α5 from cyclin against the top of the N-lobe of CDK, especially helix αC. Therefore, despite structural differences between the asymmetric dimer of the EGF receptor kinase domains and the CDK/cyclin complex, the nature of the interaction that results in activation is conceptually similar. In both cases, the adoption of the active conformation requires reorganization in the N-lobe of the kinase that leads to the exposure of hydrophobic residues (the activator-binding patch in the EGF receptor and the cyclin-binding patch in CDKs). Binding of a cyclin or the activator kinase buries these hydrophobic residues in CDK and the EGF receptor, respectively, and stabilizes the active conformation (Figure 4A and B).
CDKs and the EGF receptor represent a group of kinases that are intrinsically stable in the CDK/Src-like inactive conformation and reach their active states only upon binding of their respective external allosteric activators (Jeffrey et al., 1995; Zhang et al., 2006). In contrast, the isolated kinase domains of c-Src and Hck tend to activate spontaneously, and the associated SH2 and SH3 domains are required to stabilize their CDK/Src-like inactive conformation (Figure 2A) (Sicheri et al., 1997; Xu et al., 1997). Other kinases may fall into the spectrum of different relative stability of the CDK/Src-like conformation between these two extreme groups.
Although we have made an analogy between the asymmetric dimer of EGF receptor kinase domains and the interaction between cyclins and CDKs, there is one crucial distinction. The cyclins have high affinity for their target kinases, and in most of the cases can switch them on without additional help. In contrast, the interaction between EGF receptor kinase domains is very weak, and the kinase domains do not interact in solution. How, then, is the asymmetric dimer stabilized? One might have thought that the answer would be that ligand-induced dimerization of the extracellular domains would be the key step in bringing kinase domains together. Instead, it turns out that the segment of the receptor that connects the transmembrane helices to the kinase domains, known as the juxtamembrane segment, suffices to bring two kinase domains together in the asymmetric arrangement.
The juxtamembrane segments play essential regulatory roles in restricting the basal activity of several receptor tyrosine kinases (Hubbard, 2004). In contrast, the juxtamembrane segment of the EGF receptor is known to be necessary for ligand-dependent activation and downstream signaling (Aifa et al., 2005; Macdonald-Obermann and Pike, 2009; Thiel and Carpenter, 2007). The analysis of crystal structures of the EGF receptor and HER4 kinase domains with their juxtamembrane segments demonstrated that the conserved C-terminal segment of the juxtamembrane domain plays an important role in stabilizing the active kinase dimer (Jura et al., 2009a; Red Brewer et al., 2009; Wood et al., 2008). It does so by providing an additional interaction between the receiver kinase, which extends its C-terminal juxtamembrane segment (denoted the juxtamembrane latch) to interact with the C-lobe of the activator kinase in the asymmetric dimer (Figure 5A and 5B). This interaction is essential for ligand-dependent EGF receptor activation and potentiates dimerization between isolated kinase domains in solution (Jura et al., 2009a; Red Brewer et al., 2009).
A segment of the EGF receptor inhibitor, Mig6, contains a sequence motif that is identical to a motif in the juxtamembrane latch in the EGF receptor. Thus Mig6 may prevent juxtamembrane latch formation in addition to blocking the asymmetric dimer (Jura et al., 2009a; Zhang et al., 2007). The juxtamembrane latch binding site on the activator kinase is also occluded by the C-terminal tail of the kinase domain when the asymmetric dimer is not formed (Jura et al., 2009a), providing a mechanism for autoinhibition that is consistent with several studies (Bublil et al., 2010; Khazaie et al., 1988; Pines et al., 2010) (Figure 5C).
An intriguing observation is that the structures of the EGF receptor and HER4 kinase domains on which the understanding of the activating juxtamembrane latch interaction are based are actually both in the CDK/Src-like inactive conformation (Figure 5B). In the case of the EGF receptor kinase, the inactive conformation is a result of a mutation in the catalytic lysine residue (Lys 721 to Met) (Red Brewer et al., 2009). In the HER4 kinase, the inactive conformation is enforced by the presence of a covalently bound inhibitor (Wood et al., 2008). In spite of being in the CDK/Src-like inactive conformation, in both structures the kinases form an asymmetric dimer that largely resembles the active complex. A similar observation was made for CDK4/cyclinD1 and CDK4/cylinD3 complexes, in which CDK4 is found in the CDK/Src-like inactive conformation despite being bound to a cyclin and being autophosphorylated on the activation loop (Day et al., 2009; Takaki et al., 2009). In the case of the EGF receptor, the ability of the juxtamembrane segments to dimerize kinases even when they are in the inactive conformation might be essential for the first step of receptor activation, when two inactive receptors are brought to close proximity by a ligand.
The juxtamembrane latch by itself is not sufficient to fully activate the EGF receptor kinase, as evidenced by several studies showing that the segment N-terminal to the latch in the juxtamembrane region is also necessary for EGF receptor activation (Aifa et al., 2005; Jura et al., 2009a; Thiel and Carpenter, 2007). This region forms an amphiphatic helix and further potentiates dimerization of the isolated kinase domain of the EGF receptor in vitro (Jura et al., 2009a; Red Brewer et al., 2009). This led to a model in which the N-terminal juxtamembrane helices form a short coiled-coil dimer, which is coupled to the dimerization of the transmembrane domains of the receptor (Jura et al., 2009a). The dimerization of the transmembrane domains of the EGF receptor family of receptors upon ligand binding has been documented (Chen et al., 2009; Duneau et al., 2007; Mendrola et al., 2002) and recently visualized by NMR analysis of transmembrane helices of the EGF receptor and HER2 (Bocharov et al., 2008; Mineev et al., 2010). These structures provide a clue as to how the transmembrane domains couple dimerization of the extracellular domains to activation of the kinase domains through the cooperation of the juxtamembrane segment (Figure 5D) (Jura et al., 2009a).
HER3 lacks two important catalytic residues: the aspartate that serves as a base (Asp 813 in EGF receptor) and the glutamate in helix αC (Glu 738 in EGF receptor). HER3 has been shown to be an inactive receptor but it forms active dimers with the other members of the EGF receptor family (Guy et al., 1994; Sierke et al., 1997; Wallasch et al., 1995). The HER3-containing receptor heterodimers are critical for the activation of the phosphoinositide 3-kinase (PI3K)/AKT pathway by the EGF receptor family due to the exclusive presence of multiple binding sites for the p85 subunit of PI3K in the C-terminal tail of HER3 (Prigent and Gullick, 1994). Inhibition of HER3 phosphorylation by targeting its active dimerization partners, EGF receptor and HER2, through the use of tyrosine kinase inhibitors or therapeutic antibodies is buffered in cancer cells by HER3 overexpression (Sergina et al., 2007). This makes the inhibition of HER3-mediated signaling an important target for drug discovery.
In the allosteric mechanism for activation of the EGF receptor family members the activator kinase does not have to be catalytically active. This indicates how heterodimerization of HER3 with the active members of the EGF receptor family leads to phosphorylation of both receptors, because HER3 can take the activator position in the asymmetric dimer (Figure 6A). Sequence conservation in HER3 shows that only the activator but not the receiver interface of HER3 remains intact (Zhang et al., 2006). Biochemical studies have shown that HER3 does indeed function as an allosteric activator for other members of the EGF receptor family (Jura et al., 2009b; Monsey et al., 2010).
The crystal structures of the HER3 kinase domain show how sequence alterations prevent HER3 from becoming the receiver kinase (Jura et al., 2009b; Shi et al., 2010). In the structure, the HER3 kinase domain is in the CDK/Src-like inactive conformation, which is stabilized by a set of hydrophobic interactions that are not present in other members of the EGF receptor family (Figure 6B). In addition, there are significant conformational changes localized to the N-lobe of the HER3 kinase domain that distort the receiver interface (Jura et al., 2009b; Shi et al., 2010). These changes are primarily localized to helix αC, which in HER3 is conserved poorly relative to other HER receptors, and is partially unwound. The distinct conformation and packing of helix αC in the HER3 structure alters the receiver interface substantially.
In the crystal structures of the HER3 kinase domain nucleotide and metal ion are bound in the active site. The possibility that HER3 might actually support catalysis was examined recently (Shi et al., 2010). An autophosphorylation rate estimated to be ~1000 fold lower than for the EGF receptor kinase domain was measured for the HER3 kinase domain when it was brought to a sub-millimolar concentrations on lipid vesicles in vitro (Shi et al., 2010). It is unclear at present whether this residual activity in HER3 plays a significant role in signaling by EGF receptor family members.
Activation by binding to a hydrophobic patch in the N-lobe (the cyclin-binding patch in CDKs and the receiver interface in the EGF receptor) is a theme that is common to several kinases, including PKA (Knighton et al., 1991), extracellular signal-regulated kinase (Erk) 2 (Zhang et al., 1994), the p21-activated kinases (PAKs) (Lei et al., 2005) and the Ret receptor tyrosine kinase (Knowles et al., 2006). These kinases have a hydrophobic patch corresponding to the cyclin-binding patch, which we will refer to “helix αC patch” from now on, but rely on an intramolecular interaction for activation (Figure 7).
In these kinases, an N- (in PAKs and Ret) or C-terminal extension (in PKA and Erk2) of the kinase domain forms an extra helix, which buries and stabilizes the helix αC patch through an intramolecular interaction. A different mechanism is operative in Aurora-A kinase, in which activation requires the binding of the microtubule-associated protein TPX2 to the helix αC patch (Bayliss et al., 2003). This interaction induces a conformational change that stabilizes the active conformation of Aurora-A kinase (Figure 7). The tyrosine kinase Fes utilizes its own SH2 domain to stabilize the “swung-in” position of helix αC in the active conformation, through interactions at an analogous hydrophobic patch (Filippakopoulos et al., 2008). A similar interaction might also play a role in the activation of the Abl tyrosine kinase, as suggested by small angle X-ray scattering analysis and mutagenesis (Filippakopoulos et al., 2008; Nagar et al., 2006).
Activation through binding of the helix αC patch is also a shared mechanism of the subset of the AGC superfamily of kinases (Pearce et al., 2010), including the protein kinase B (PKB/Akt) (Yang et al., 2002), Rho-kinase (ROCK) (Yamaguchi et al., 2006), the phosphoinositide-dependent protein kinase 1 (PDK1) (Biondi et al., 2002) and protein kinase C (PKC) (Grodsky et al., 2006). In these kinases the helix αC patch is denoted the hydrophobic motif-binding pocket (HM/PIF binding pocket). In the crystal structures of PKB/Akt, PKC and ROCK, the HM-binding pocket is bound to a phosphorylated hydrophobic motif (HM) in the C-terminal extensions of these kinases, and this stabilizes the active conformation (Figure 7). In the Rho-kinase the HM motif itself is stabilized by hydrophobic interactions with the N-terminal extension that forms a helix bundle to mediate dimerization of the kinase (Figure 7). In the absence of phosphorylation, the HM motif does not bind to the helix αC patch, and these AGC kinases are inactive (Pearce et al., 2010).
The HM/PIF binding pocket plays a dual role in regulation of the PDK1 kinase, which activates many AGC kinases through phosphorylation of their activation loops. PDK1 uses its HM/PIF binding pocket to dock the substrate AGC kinases by using their HM motifs (Biondi et al., 2000). The docking interaction also results in the stabilization of the HM/PIF binding pocket of PDK1 and allosteric activation of PDK1 activity. Small molecule compounds have been developed that bind to the HM/PIF binding pocket in PDK1 and induce PDK1 activation (Engel et al., 2006; Hindie et al., 2009). The discovery of these compounds opens a new chapter in the development of drugs targeting the AGC kinases, by uncoupling PDK1 activity from its binding to the substrate AGC kinases. Such compounds may also represent a general class of inhibitors that target kinases, including the EGF receptor family, whose activation is dependent on the helix αC patch. These inhibitors are likely to be more specific than the ATP analogs since they target a much more diverse interface than the highly conserved nucleotide binding site. In case of the EGF receptor family, the idea of targeting the helix αC patch is particularly exciting as a strategy to uncouple the catalytically inactive HER3 from its active dimerization partners.
The CDK/Src-like inactive conformation, which is observed in structures of several unrelated kinases, emerges as an essential and possibly conserved step in kinase catalysis. It is possible that other inactive conformations that have been observed in crystal structures also represent important intermediates in the kinase catalytic cycle and they should remain an important subject of structural investigation.
One important aspect of the analysis of kinase inactive states is that it can also shed light on the mechanisms of kinase activation. This was important for the elucidation of the activation mechanism of the EGF receptor family of tyrosine kinases, which turns out to be very different from the way other receptor tyrosine kinases are activated and more similar to how CDKs are activated by cyclins. The analogy between the activation of CDKs and the EGF receptor family also underscores how distantly related kinases, with diverse activation mechanisms, can share common structural mechanisms. In fact, these commonalities might be more general and our analysis of different kinase structures demonstrates that a hydrophobic patch in the N-lobe of the kinase (the helix αC patch) serves as a binding site for activators in kinases representing different branches on the kinome tree. The detailed analysis of these common themes and intrinsic differences is crucial for the advancement of our understanding of how different kinases activate and how we can explore these specific differences to design selective strategies for inhibiting aberrant activation of these kinases in human disease.