Autoregulatory Mechanisms in Protein-tyrosine Kinases*

Protein-tyrosine kinases (PTKs), enzymes that catalyze the transfer of the g-phosphate of ATP to tyrosine residues of protein substrates, are critical components of signaling pathways that control cellular proliferation and differentiation. PTKs can be subdivided into two large families, receptor tyrosine kinases (RTKs) and non-receptor tyrosine kinases (NRTKs) (1, 2). RTKs span the plasma membrane and contain an extracellular portion, which binds ligand, and an intracellular portion, which possesses catalytic activity and regulatory sequences. The RTK family includes the insulin receptor and the receptors for many growth factors such as epidermal (EGF), platelet-derived (PDGF), fibroblast (FGF), and nerve growth factors (1). NRTKs contain no extracellular or transmembrane portion but possess modular domains that are responsible for subcellular targeting and regulation of catalytic activity. The NRTK family includes Src, Abl, FAK, and the JAKs among many others (2). Because of the key roles PTKs play in cellular signaling processes, their catalytic activity is tightly controlled in normal cells by protein-tyrosine phosphatases, by other protein tyrosine or serine/threonine kinases (1), and by autoregulatory mechanisms. The recent crystallographic structures of several members of both the RTK and NRTK families, together with extensive biochemical studies, afford an understanding at the molecular level of the autoregulation mechanisms to which PTKs are subject.


Receptor Tyrosine Kinases
The RTK family can be broadly divided into two groups depending on the covalent organization of the receptor. Most RTKs possess a single polypeptide chain and are monomeric in the absence of ligand. Members of the insulin receptor subfamily, which includes insulin-like growth factor-1 receptor, are disulfide-linked dimers of two polypeptide chains, forming an ␣ 2 ␤ 2 heterotetramer. Ligand binding to the extracellular portion of RTKs leads to dimerization of monomeric receptors or a rearrangement within the quaternary structure of heterotetrameric receptors, resulting in autophosphorylation of specific tyrosine residues in the cytoplasmic portion (1).
In general, tyrosine autophosphorylation either stimulates the intrinsic catalytic (kinase) activity of the receptor or gen-erates recruitment sites for downstream signaling proteins containing phosphotyrosine-recognition domains, such as the Src homology 2 (SH2) domain or the phosphotyrosine-binding (PTB) domain (3). The vast majority of PTKs contain between one and three tyrosines in the kinase activation loop (A-loop), which comprises subdomains VII and VIII of the protein kinase catalytic core (4). Phosphorylation of these tyrosines has been shown to be critical for stimulation of catalytic activity and biological function for RTKs such as the insulin receptor (5), FGF receptor (6), hepatocyte growth factor receptor (MET) (7), and nerve growth factor receptor (TrkA) (8), and for NRTKs such as Src (9), Zap-70 (10), and JAK2 (11).
Efficient phosphorylation of protein substrates by RTKs generally requires not only stimulation of receptor catalytic activity but also localization of protein substrates to the activated receptor via a physical association extending beyond a simple enzyme-substrate interaction. For example, autophosphorylation of Tyr 766 in FGF receptor 1 provides a high affinity binding site for the SH2 domain of phospholipase C␥. The association of the SH2 domain with phosphorylated Tyr 766 (pTyr 766 ) is critical for the phosphorylation and activation of phospholipase C␥ by FGF receptor 1 (12). Similarly, the PTB domain of insulin receptor substrate 1 (IRS-1) is recruited to pTyr 972 of the activated insulin receptor, while the pleckstrin homology domain appears to target IRS-1 to the plasma membrane (13). Both of these interactions serve to localize IRS-1 to the activated insulin receptor, permitting phosphorylation of numerous tyrosines in IRS-1.

Cis-inhibition/Trans-activation
Crystal structures of the unphosphorylated forms of the insulin receptor kinase domain (IRK) (14) and the FGF receptor 1 kinase domain (FGFR1K) (15) have provided details on the molecular mechanisms by which RTKs are kept in a low activity state prior to autophosphorylation of A-loop tyrosines. In the IRK structure, one of the three tyrosines in the kinase A-loop, Tyr 1162 , is bound in the active site, seemingly in position to be autophosphorylated (in cis) (Fig. 1A). However, Asp 1150 of the protein kinase-conserved Asp-Phe-Gly sequence motif at the beginning of the A-loop is not in proper position to coordinate MgATP (16,17); in fact, the beginning of the A-loop appears to interfere with ATP binding. Biochemical data are consistent with autophosphorylation of Tyr 1162 (and Tyr 1158/1163 ) occurring in trans (by a second IRK molecule) (18). Moreover, substitution of Tyr 1162 with phenylalanine in the intact insulin receptor results in an increase of basal (absence of insulin) kinase activity (5), consistent with an autoinhibitory role for Tyr 1162 . Thus, the available evidence suggests that prior to autophosphorylation, Tyr 1162 competes with protein substrates (neighboring ␤ chain and other substrates) for binding in the active site but is not cis-autophosphorylated because of steric constraints that prevent simultaneous binding of Tyr 1162 and MgATP.
Although the A-loop sequences of the insulin receptor and FGFR1 are ϳ50% identical, the A-loop as observed in the FGFR1K crystal structure adopts a significantly different conformation than that observed in IRK (15). In FGFR1K, neither of the two A-loop tyrosines, Tyr 653 and Tyr 654 (Tyr 1162/1163 in IRK), is bound in the active site. Instead, the PTK-invariant proline at the end of the A-loop, Pro 663 (Pro 1172 in IRK), and residues N-terminal are positioned to interfere with binding of protein substrates (Fig. 1B). Because of the difference in conformation of the A-loop near the Asp-Phe-Gly motif, the ATP binding site in FGFR1K is not obstructed, and indeed ATP can be soaked into crystals and observed to bind. This conformational difference may in part be due to the residue that immediately precedes Asp-Phe-Gly, which is a glycine (Gly 1149 ) in IRK and an alanine (Ala 640 ) in FGFR1K. Due to steric hindrance, it appears that only a glycine at this position allows the polypeptide chain to traverse the ATP binding cleft between the N-and C-terminal lobes of the kinase (Fig. 1A).
The structures of tris-phosphorylated IRK (17) and monophosphorylated Lck (19) provide a molecular basis for understanding how autophosphorylation of A-loop tyrosines stimulates catalytic activity. Autophosphorylation of the three tyrosines in the IRK A-loop results in a dramatic change in the configuration of the loop (Fig. 1A). Stabilization of this A-loop conformation involves both phosphotyrosine and non-phosphotyrosine interactions (17,19). The conformation of the phosphorylated A-loop permits unrestricted access to the binding sites for ATP and protein substrates and facilitates the proper spatial arrangement of residues involved in MgATP coordination, namely the protein kinase-conserved lysine and glutamic acid from the N-terminal lobe (Lys 1030 and Glu 1047 in IRK) and the aspartic acid of the conserved Asp-Phe-Gly triad (Asp 1150 in IRK).
Why is some form of autoinhibition necessary for the kinase domain of RTKs? The kinase domains of the insulin and insulin-like growth factor-1 receptors are always within close proximity, and autoinhibition would serve to minimize the extent of ligand-independent autophosphorylation. Evidently for monomeric RTKs in the absence of ligand, random collisions within the plane of the plasma membrane between non-inhibited receptor molecules would be sufficient to result in an appreciable amount of autophosphorylation and hence activation. The insulin receptor (and most RTKs) can be activated in the absence of ligand by tyrosine phosphatase inhibitors such as vanadate, which can mimic some of the biological effects of insulin (20,21), indicating that autoinhibition alone is not sufficient to keep RTKs quiescent.
It is clear from the crystallographic temperature factors (Bfactors) that segments of the unphosphorylated IRK and FGFR1K A-loops are relatively mobile, and therefore an equilibrium between different conformations of the A-loop likely exists in vivo. A subset of these conformations will be compatible with substrate binding, and in fact substrates (protein and ATP) will compete for binding with the A-loop. Phosphorylation of tyrosine(s) in the A-loop will markedly shift the equilibrium toward a conformation that accommodates substrate binding. It seems plausible that the equilibrium for a particular RTK has been "tuned" to provide inhibition strong enough to deter phosphorylation of substrates (receptor or other proteins) in the absence of ligand yet weak enough to permit trans-autophosphorylation of receptors that have been juxtaposed via ligand binding.
Activating point mutations in the A-loops of various RTKs have been implicated in human disease. Mutations of an aspartic acid after the Asp-Phe-Gly motif in Kit and MET have been identified in patients with mast cell leukemia and papillary renal carcinoma, respectively (22,23). In FGF receptor 3, a substitution of glutamic acid for lysine (Lys 650 -Glu) following the tandem tyrosines in the A-loop underlies a form of thanatophoric dysplasia (24), a lethal skeletal dysplasia. Compared with wild type, all of these mutant receptors are heavily tyrosine-phosphorylated when expressed in cultured cells in the absence of ligand (25,26). Presumably, these mutations shift the above mentioned A-loop equilibrium toward the active conformation.

Dimerization Mechanisms
Binding of a dimeric ligand such as PDGF to its receptor likely induces the formation of a symmetric dimer of the receptor extracellular domains. The recent crystal structure of vascular endothelial growth factor bound to immunoglobulin-like domain 2 (D2) of its receptor Flt-1 provides an example (27). The structure shows a 2-fold symmetric arrangement of vascular endothelial growth factor, which is a disulfide-linked dimer, and two molecules of D2. Less clear is the relative disposition of the cytoplasmic domains in the receptor pair upon ligand binding. If a stable symmetric (or asymmetric) cytoplasmic dimer is formed, autophosphorylation of at least some sites would presumably have to occur between dimers rather than within dimers because of steric constraints (15). The functional utility of a cytoplasmic dimer could be to stabilize a more active state of the kinase. Although a lone tyrosine in the A-loop is conserved throughout the EGF receptor subfamily, substitution of Tyr 845 with phenylalanine has no demonstrable effect on EGF receptor kinase activity or signaling properties (28). In this case, stimulation of kinase activity upon EGF binding may arise from formation of a cytoplasmic dimer in which the Aloops are stabilized in a conformation favorable for catalysis.
Alternatively, the cytoplasmic domains of the receptor pair may associate only transiently, the two cytoplasmic domains acting simply as substrate and enzyme for one another. The substantial increase in the local substrate/enzyme concentration afforded by receptor dimerization (extracellular) could provide sufficient opportunity for trans-autophosphorylation to occur between partially autoinhibited kinases (Fig. 2). When insulin is removed from the activated insulin receptor, the increased kinase activity of the receptor is maintained (29), suggesting that once autophosphorylation (and A-loop rearrangement) occurs, a particular spatial arrangement of the cytoplasmic domains is not necessary to sustain activity.

Non-receptor Tyrosine Kinases
In addition to a tyrosine kinase domain, NRTKs often contain within the same polypeptide chain several protein-protein or protein-lipid interaction modules such as SH2, SH3, and pleckstrin homology domains. For example, after a myristoylation site and a unique N-terminal region, Src contains an SH3 domain followed by an SH2 domain, a kinase domain, and a short C-terminal segment. As will be discussed below, extracellular stimuli lead to the activation of NRTKs by both intramolecular and intermolecular mechanisms.
Biochemical studies, as well as an oncogenic form of Src (v-Src), had suggested that Src was autoinhibited through interaction of its own SH2 domain with a phosphotyrosine (pTyr 527 ) just C-terminal to the kinase domain (30,31), which is phosphorylated by the NRTK Csk (32). The SH3 domain was also known to be involved in regulation of catalytic activity and oncogenesis (2), but the actual mechanism by which this occurs was not understood. The recent crystal structures of nearly full-length Src (33,34) and Hck (35) have elucidated the roles of the SH2 and SH3 domains in the autoregulation of Src family kinases.
In the structures of Src and Hck, the SH2 domain does indeed interact with pTyr 527 in the C-terminal tail as had been predicted. This interaction by itself, however, would not appear to inhibit kinase activity because the interaction occurs on the back side of the C-terminal lobe of the kinase, away from the active site. The most striking feature in the structures is the interaction of the SH3 domain with a portion of the segment that links the SH2 domain to the kinase domain (Fig. 3). SH3 domains are known to bind to proteins containing Pro-X-X-Pro sequences that form polyproline type II helices (36). Within the SH2-kinase linker of Src family members such as Hck and Lck, a Pro-X-X-Pro motif is present. But for other family members like Src and Fyn, the second proline is absent, and therefore, an interaction between the SH3 domain and the SH2-kinase linker was not anticipated. Despite the absence of the second proline in Src, this portion of the SH2-kinase linker forms a polyproline type II helix to which the SH3 domain binds.
How do the SH2 and SH3 domains repress kinase activity? Helix C in the N-terminal lobe of protein kinases contains an invariant glutamic acid (Glu 310 in Src) that in active kinases forms an ion pair with an invariant lysine from ␤ strand 3 (Lys 295 in Src) (19,37). In the Src and Hck structures, however, the distance between Glu 310 and Lys 295 is greater than 12 Å. The position of helix C is influenced by interactions between the kinase N-terminal lobe and the SH2-kinase linker, whose position in turn is fixed by the SH3 domain (33)(34)(35). Repression of kinase activity is likely due to the displacement of helix C as well as a restriction of the relative motion between the N-and  (33). B, the kinase domain of Src is held in an inactive state through two distinct intramolecular interactions: the binding of the SH2 domain to pTyr 527 and the binding of the SH3 domain to the SH2-kinase linker, which contains a short polyproline type II helix. As a result of extracellular stimuli, the kinase activity of Src can be enhanced by binding of the SH3 domain to proline-rich sequences, binding of the SH2 domain to phosphotyrosinecontaining sequences, or dephosphorylation of pTyr 527 . Full activation of Src requires autophosphorylation of Tyr 416 in the A-loop.
C-terminal lobes of the kinase, which must open and close to bind ATP and release ADP.
Src is activated in response to growth factors such as PDGF, which induces PDGF receptor dimerization and autophosphorylation of numerous tyrosines. An autophosphorylation site in the juxtamembrane region of the PDGF␤ receptor (pTyr 579 ) can bind the SH2 domain of Src with high affinity (38,39), effectively competing with the pTyr 527 of Src in the C-terminal tail. Binding of the Src SH2 domain to pTyr 579 of the PDGF␤ receptor leads to stimulation of Src kinase activity by releasing the intramolecular constraints on the kinase domain (Fig. 3B). Alternatively, Src can be activated by proteins containing proline-rich sequences that effectively compete with the SH2-kinase linker for binding to the SH3 domain (2,40), which again results in disruption of the inhibitory intramolecular constraints. Once released from the autoinhibited state, Src undergoes trans-autophosphorylation on a conserved tyrosine residue in the A-loop (Tyr 416 ), which stabilizes the active A-loop conformation (19).
Members of the JAK family of NRTKs (JAK1-3 and TYK2) are bound to the cytoplasmic domain of lymphokine receptors through non-covalent interactions. Binding of a lymphokine to the extracellular domain of its receptor leads to receptor dimerization and the juxtaposition of two JAKs (which may be different family members) (41). The mechanism of activation of JAKs by lymphokines appears to be very similar to that of RTKs by their specific ligands. In both cases, receptor dimerization increases the local concentration of catalytic domains, which are either intrinsic (RTKs) or extrinsic (JAKs), enabling trans-phosphorylation of A-loop tyrosines. These phosphotyrosines function to maintain an active kinase state until dephosphorylation by tyrosine phosphatases restores the low activity, basal state.

Conclusions
The crystal structures of the catalytic domains of the insulin receptor, FGF receptor, and Lck, and the nearly full-length structures of Src and Hck, have provided a wealth of information concerning the mechanisms by which tyrosine kinase activity is regulated. We have gained an understanding of how tyrosine autophosphorylation stimulates catalytic activity and of how the various domains of the Src family PTKs cooperate to suppress catalytic activity. Moreover, the structures of PTKs in complex with small molecule inhibitors (42) should lead to the generation of more potent and specific inhibitors for the treatment of diseases such as cancer, in which aberrant activity of various PTKs has been implicated (43). Future structural and functional studies of other receptor and non-receptor PTKs will likely unveil other autoregulatory mechanisms to which this important class of enzymes is subject.