|Home | About | Journals | Submit | Contact Us | Français|
CASK is a unique MAGUK protein that contains an N-terminal CaM-kinase domain besides the typical MAGUK domains. The CASK CaM-kinase domain is presumed to be a catalytically inactive pseudokinase because it lacks the canonical DFG-motif required for Mg2+-binding. Here we show, however, that CASK functions as an active protein kinase even without Mg2+. High-resolution crystal structures reveal that the CASK CaM-kinase domain adopts a constitutively active conformation that binds ATP and catalyzes phosphotransfer without Mg2+. The CASK CaM-kinase domain phosphorylates itself and at least one physiological interactor, the synaptic protein neurexin-1, to which CASK is recruited via its PDZ-domain. Thus, our data indicate that CASK combines the scaffolding activity of MAGUKs with an unusual kinase activity that phosphorylates substrates which are recuited by the scaffolding activity. Moreover, our study suggests that besides CASK, other pseudokinases (10% of the kinome) could also be catalytically active.
At least 1.7% of human genes encode protein kinases (Manning et al., 2002). 10 % of all kinase domains have accumulated evolutionary changes that, according to our present understanding, render them catalytically inactive, prompting these to be designated “pseudokinases” (Manning et al., 2002). Sequence analyses of these domains revealed evolutionarily conserved alterations in otherwise highly invariant motifs. For instance, the Ca2+/calmodulin-dependent (CaM)-kinase domain of CASK (Ca2+/calmodulin activated Ser-Thr kinase) is one of nine known pseudokinases that contain alterations in the canonical Mg2+-binding DFG (Asp-Phe-Gly) motif and are thought to have no catalytic activity (Boudeau et al., 2006; Manning et al., 2002).
CASK was simultaneously discovered in three different organisms: In biochemical experiments in vertebrates because of its binding to neurexins, which are synaptic cell-adhesion molecules (Hata et al., 1996); in genetic experiments in C. elegans because it is encoded by the Lin-2 gene that is essential for vulva development (Hoskins et al., 1996); and by sequence analyses in D. melanogaster because CASK contains an N-terminal CaM-kinase domain (Martin and Ollo, 1996). Deletion or mutation of CASK leads to anomalous synaptic function and perinatal death in mice (Atasoy et al., 2007), to severe developmental deficits in C. elegans (Hoskins et al., 1996), or to behavioral and neurotransmitter-release abnormalities in Drosophila (Lopes et al., 2001; Martin and Ollo, 1996; Zordan et al., 2005). In humans, the CASK gene has been linked to X-linked optic atrophy and mental retardation in mapping and cohort studies (Dimitratos et al., 1998; Froyen et al., 2007). The evolutionary conservation of CASK and the severe deletion phenotypes suggest that CASK performs important biological roles, but the precise nature of its function has remained elusive.
CASK is composed of an N-terminal CaM-kinase domain that accounts for approximately half of its primary structure, and a C-terminal combination of domains that is typical of MAGUKs (membrane-associated guanylate kinases). The MAGUK-half of CASK includes an L27 domain, a PDZ domain (which was one of the first PDZ-domain structures determined (Daniels et al., 1998)), an SH3-domain, and a C-terminal guanylate kinase domain that is thought to engage in inter- and intramolecular interactions (Hsueh et al., 2000; Nix et al., 2000).
Since the CASK CaM-kinase domain is presumed inactive, studies of CASK function have focused on its role as a MAGUK-type scaffolding protein. A large array of CASK interactions has been described: CASK associates stoichiometrically with Mint-1 (also called X11 or Lin-10) and Velis (also called MALS or Lin-7) that may be involved in organizing synapses (Borg et al., 1998; Butz et al., 1998). Consistent with a scaffolding activity, the CASK PDZ-domain binds to neurexins, syndecans, and SynCAMs, which are putative cell-adhesion molecules (Biederer et al., 2002; Cohen et al., 1998; Hata et al., 1996; Hsueh et al., 1998). CASK also associates with protein-4.1, and a trimeric complex composed of CASK, protein-4.1, and the cytoplasmic tail of neurexin-1 is a potent nucleator of actin polymerization (Biederer and Sudhof, 2001). CASK may recruit Ca2+ channels to the synapse (Maximov et al., 1999), and target potassium channels (Leonoudakis et al., 2004) and/or the Ca2+ pump 4b/Cl (Schuh et al., 2003) to the plasma membrane. In C. elegans, the CASK homolog Lin-2 is implicated in proper localization of the EGF-receptor LET-23 (Hoskins et al., 1996). CASK is proposed to bind PARKIN and function in its targeting and scaffolding at the synapse (Fallon et al., 2002). Similarly, the CASK-Mint-1 complex has been implicated in targeting the NMDA-receptor subunit NR2B from Golgi to postsynapse via its interaction with dendritic kinesin KIF17 (Setou et al., 2000). CASK also associates with CASKins (Olsen et al., 2005) and liprins (Setou et al., 2000), and might regulate transcription factors in the nucleus (Hsueh et al., 2000). Moreover, CASK enhances Ether-a-go-go potassium currents via increasing its phosphorylation (Marble et al., 2005). Finally, CASK may associate with CaM-kinase II (CaMKII), and modulate its phosphorylation state (Hodge et al., 2006; Lu et al., 2003; Zordan et al., 2005). The regulation of phosphorylation was thought to be indirect in both the cases, as CASK was deemed a pseudokinase.
Here, we decribe high-resolution crystal structures of the CaM-kinase domain of CASK complexed to different nucleotides. These structures led us to hypothesize that despite the lack of an appropriate Mg2+-binding motif in its structure, CASK may nevertheless function as a protein kinase. Indeed, we show that unlike any known kinase, the CASK CaM-kinase domain catalyzes autophosphorylation and phosphotransfer to the cytoplasmic tail of neurexin-1 in the absence of Mg2+. Our data suggest that CASK is unique among protein-kinases in that its CaM-kinase domain uses an unusual mechanism to phosphorylate substrates which are recruited via its MAGUK component.
We compared the sequences of the CASK CaM-kinase domain with the homologous kinases CaMKI (≈ 37 % identity) and CaMKII (≈ 44 % identity; Suppl. Fig. S1). Based on the sequence homology, we produced a recombinant CASK CaM-kinase domain (residues 1-337; note that these residues are identical in mouse, rat and human CASK), and determined its crystal structure. The purified protein was crystallized in two space groups, a reticularly twinned triclinic form (P1 form) and an orthorhombic form (P212121 form). The structures of both forms were solved by molecular replacement using the coordinates of the CaMKI kinase domain (Goldberg et al., 1996) (PDB ID: 1A06), and refined at 2.2 Å (P1 form) and 1.85 Å (P212121 form) resolution (Fig. 1, Suppl. Fig. S3, and Suppl. Table S1). The structures of the CASK CaM-kinase domain in the two crystal forms are highly similar (pairwise Cα root-mean-square deviation ≈ 0.4 Å), indicating that the overall structure of the domain was not significantly influenced by the crystalline environment (Figs. 1A,B, Suppl. Figs. S2A and S3). Both the P1 and P212121 structures exhibited uninterrupted backbone electron densities throughout the entire domain from residues 5 to 304.
The catalytic core of protein kinases can be divided into an N-terminal lobe that is involved in ATP binding, and a C-terminal lobe that binds and positions the substrate in addition to contributing to ATP binding and ATP activation (Taylor et al., 2004). In the active state conformation, protein kinases adopt a very similar overall fold that positions the elements in the N- and C-terminal lobes involved in substrate binding and catalysis in a canonical fashion (Huse and Kuriyan, 2002; Nolen et al., 2004). In the inactive state, in contrast, protein kinases adopt different overall folds.
The CASK CaM-kinase domain exhibits a typical protein kinase fold with an N-terminal lobe dominated by a five-stranded β-sheet and an α-helical C-terminal lobe (Figs. 1A,B, and Suppl. Figs. S2 and S3). Comparison of the CASK CaM-kinase domain structures with those of other protein kinases revealed that the relative orientation of N- and C-terminal lobes in the CASK CaM-kinase domain is very similar to that of the active state of the death-associated protein kinase 1 (DAPK1) in a complex with the ATP analog AMPPNP (Tereshko et al., 2001) (PDB ID: 1IG1), resembling the catalytically active “closed conformation” (Figs. 1A,B, and D, and Suppl. Fig. S2C). In contrast, the inactive form of CaMKI adopts a distinct “open conformation” with respect to the two lobes, resulting in a diminished affinity for nucleotides (Figs. 1C and Suppl. Fig. S2B) (Goldberg et al., 1996), that is very different from that of the CASK CaM-kinase domain.
Equivalents of all the essential elements were found in this domain (Figs. 1A,B, and Suppl. Figs. S2 and S3). The N-terminal lobe encompasses a Gly-rich loop (residues 19–24 in CASK; brown) and a Lys-Glu ion pair (Lys41 and Glu62 in CASK) (Suppl. Fig. S5C). Furthermore, it displays an activation segment (green) and a catalytic loop (residues 141–146 in CASK; yellow). This catalytic loop is positioned normally in CASK CaM-kinase domain, although the invariant Asn146 is substituted by Cys (Suppl. Fig. S1). In protein kinases, Asn146 in the catalytic loop coordinates Mg2+ and stabilizes the catalytic loop by forming a hydrogen bond with the catalytic Asp141. In the absence of an Asn at position 146, the catalytic Asp141 of the CASK CaM-kinase domain engages in a number of alternative interactions, which stabilize its conformation and the catalytic loop. Notably, the smaller Cys146 side chain is still able to engage in a hydrogen bond with the backbone carbonyl of Asp141 (Suppl. Fig. S5A,B).
In kinase structures corresponding to an inactive state, such as the structures of CaMKI, MnK1 and MnK2, the activation segment is often displaced from its binding site in the C-terminal lobe, and frequently is disordered (Fig. 1C and Suppl. Fig. S2B) (Goldberg et al., 1996; Jauch et al., 2006; Jauch et al., 2005). However, in the CASK CaM-kinase domain, the activation segment is fully ordered, and its C-terminal portion is cradled in an active state conformation within the C-terminal lobe. Superimposing CDK2 in complex with cyclin A, a substrate peptide and ATP (PDB ID 1QMZ) with the CASK structure suggests that substrate peptides can bind on top of the activation segment of the CASK CaM-kinase domain along the mouth of the nucleotide binding cleft. ATP thus could be bound next to the substrate into the nucleotide binding pocket with its γ-phosphate pointing towards the substrate peptide, in a fully active orientation (Suppl. Fig S4B).
The C-terminal lobe of the CASK CaM-kinase domain contains an equivalent of the Mg2+-binding loop, but the DFG-motif within the loop is altered to GFG by the Asp162Gly substitution. This loop is ordered in the CASK CaM-kinase domain structure (orange) as in DAPK1, while it is partially disordered in the CaMKI structure (Fig. 1C and S2B). Overall, the potential catalytic elements of the CASK CaM-kinase domain could be superimposed onto corresponding elements in the DAPK1-AMPPNP co-crystal structure (Suppl. Fig. S2C), confirming that the structures exhibit core features and functional elements closely resembling an inherently active-state fold; dissimilar to the inactive CaMKI fold.
The nucleotide binding pocket of the CASK domain was occupied by 3'-AMP in the crystal structure (Figs. 1A and and2D,2D, and Suppl. Figs. S2A and S3, and Suppl. Table S1). The bound 3'-AMP is likely a byproduct of bacterial RNA degradation during protein purification. Upon soaking the crystals in 5'-AMPPNP in the presence of equimolar Mg2+, the 3'-AMP could be displaced by 5'-AMPPNP, suggesting a readily accessible adenine-nucleotide binding site (Figs. 1B, 2B, and 2E, and Suppl. Fig. S2A). The CASK CaM-kinase domain, thus, adopts an intrinsically active conformation amenable to nucleotide binding.
To better characterize the mode of nucleotide binding to the CASK CaM-kinase domain, we solved the structure of the best derivative crystals containing 5'-AMPPNP at 2.3 Å resolution (P1 form; Suppl. Table S1). The binding mode of the nucleotide in the pocket is reminiscent of the AMPPNP-bound DAPK1 structure (Tereshko et al., 2001). The N1 and N6 amino groups of the adenine form hydrogen-bonds with the backbone NH of Met94 and the oxygen of Glu92, respectively. Similar to the orientation of AMPPNP in the DAPK1 structure, the adenine base of the 5'-AMP moiety was found in an anti conformation, and the sugar retained a C2'-endo conformation (Fig. 2B). The overall binding and orientation of the 5'-AMP portion of the nucleotide is in agreement with the binding of ATP to generic protein kinases in an active conformation (Figs. 2A–2E).
In active protein kinases, bound nucleotides are coordinated by a Mg2+ ion and by amino acid residues in the surrounding pocket. DAPK1 coordinates the α-phosphate of AMPPNP through a hydrogen bond with Lys42. In addition, a Mg2+ ion (that in turn is bound to the Asp of the DFG-motif and Asn of the catalytic loop) bridges the α- and β-phosphates (Tereshko et al., 2001) (Fig. 2C). In the CASK CaM-kinase structure containing AMPPNP, the α-phosphate was resolved and found to be positioned between the side chains of His145 and Lys41, without directly contacting either residue. Rather, the α-phosphate is heavily hydrated and maintains water-mediated contacts to various sites in the pocket (Fig. 2B and Suppl. Fig. S5C). The β- and γ-phosphates, in contrast, were disordered; no Mg2+ ion could be discerned in the pocket, despite its presence in the crystallizing milieu. With the lack of a canonical Mg2+ binding motif in the CASK CaM-kinase domain, the phosphates may be de-coordinated by the coupled Mg2+.
Thus, the structural data suggest that the CASK CaM-kinase domain is capable of coordinating ATP in a canonical orientation without Mg2+, and is equipped with all the structural components of a catalytic kinase, with noticeable substitutions in both the Mg2+-coordinating residues (Asp162Gly and Asn146Cys) and a consequential de-coordination of the β- and γ-phosphates of ATP. These results indicate that the CASK CaM-kinase domain may physiologically bind nucleotides without Mg2+-binding. To explore this hypothesis, we first studied nucleotide binding to the CASK CaM-kinase domain.
Since the crystal structure showed that the CASK CaM-kinase domain binds ATP, we tested whether the domain interacts with the ATP analog TNP-ATP, that becomes fluorescent when inserted into the hydrophobic ATP-binding pocket (Stewart et al., 1998).
In the presence of EDTA, the CASK CaM-kinase domain induced a 3 to 5-fold increase in fluorescence intensity, and a blue-shift of the emission maximum, of TNP-ATP (Fig. 3A). ATP partly blocked this fluorescence increase. Titrations revealed that a 500-fold excess of ATP reduced the TNP-ATP fluorescence by 50% (KiATP ≈ 0.365 mM), consistent with competitive binding of ATP and TNP-ATP to the same site, that exhibits a much higher affinity for TNP-ATP than for ATP (Stewart et al., 1998). Even a 1,000-fold excess of GTP had no effect on TNP-ATP binding, pointing to adenine nucleotide-specificity of the pocket (Fig. 3B). To further dissect the specificity of nucleotide binding to the CASK CaM-kinase domain, we examined the inhibition of TNP-ATP binding by adenine, ATP, ADP, 5'-AMP, 3'-AMP, and cAMP. Only ATP inhibited TNP-ATP binding to the CASK CaM-kinase domain efficiently, while all other naturally occurring nucleotides tested had little effect (Fig. 3C), suggesting that the CASK CaM-kinase domain specifically binds ATP.
Although certain kinases can utilize Mn2+ for enzymatic activity in vitro (Nirmala et al., 2006), Mg2+ is the only divalent ion available at optimal concentration in vivo. Thus, kinases have evolved to exclusively utilize Mg2+ as a cofactor for catalysis (Waas et al., 2004). We tested the effect of divalent cations on TNP-ATP binding to the CASK CaM-kinase domain. Surprisingly, addition of Mg2+ inhibited the increase in TNP-ATP fluorescence induced by the CASK CaM-kinase domain (Fig. 3A). Titration experiments demonstrated that the TNP-ATP – CASK CaM-kinase domain complex is diminished at higher than equimolar concentrations of Mg2+ (KiMg2+ ≈ 10.41 μM), suggesting that Mg2+ competitively inhibits ATP-binding to the CASK CaM-kinase domain (Suppl. Fig. S6B). Mn2+ and Ca2+ could also inhibit TNP-ATP binding (Fig. 3D). The TNP moiety of TNP-ATP could potentially form atypical contacts in the binding site, as demonstrated for the histidine kinase CheA (Bilwes et al., 2001), and TNP-ATP binding may not be representative of ATP binding. To address this possibility, we tested binding of the CASK CaM-kinase domain to γ32P-ATP using a dot-blot assay (Suppl. Fig. S6A). This assay revealed that Mg2+ significantly inhibited binding of γ32P-ATP to the CASK CaM-kinase domain. Hence, ATP binding to the CASK CaM-kinase domain is disrupted by biologically abundant divalent cations, including Mg2+, consistent with the absence of a DFG-motif and a C-loop Asn in its structure.
Coordination of the β- and γ-phosphates of ATP by Mg2+-ions catalyzes phosphotransfer reactions by protein kinases (Adams, 2001; Hanks and Hunter, 1995). The fact that Mg2+, if anything, impairs ATP binding by the CASK CaM-kinase domain is consistent with the notion that this domain is indeed a `pseudokinase'; despite its constitutively active conformation and avid ATP binding. To test this, we examined whether the CASK CaM-kinase domain exhibits autophosphorylation activity, which is typical for CaM-kinases and can be employed as a test for kinase activity (Hanley et al., 1988).
Strikingly, the CASK CaM-kinase domain was autophosphorylated with nearly 13 % efficiency (Fig. 4A). Autophosphorylation was independent of Mg2+, since it was carried out in Mg2+-free buffer containing 2 mM EDTA. Tandem mass-spectrometry identified a single autophosphorylated CASK peptide containing residues 141–159 (DVKPHCVLLASKENSAPVK; data not shown). The phosphorylated peptide is located in a surface-exposed loop preceding the GFG motif (Suppl. Fig. S4A). Since it is remote from the substrate binding pocket, we speculate that autophosphorylation occurs in trans.
The autophosphorylation assay was used to calculate the Michaelis constant of the CASK CaM-kinase domain for ATP (KmATP) (Fig. 4B), revealing a KmATP of ≈ 0.563 mM, which is within the range of KmATP for active kinases (Warmuth et al., 2007). We measured the limiting rate of autophosphorylation in experiments that approached saturation. Again, the Vmax for autophosphorylation of the CASK CaM-kinase domain (≈ 4.9 nmol/μmol enzyme/min), although low, is similar that of some other kinases, such as Giardia lamblia PKA (Vmax ≈ 6.2 nmol/μmol enzyme/min) (Abel et al., 2001).
Mg2+ and other divalent cations inhibited but did not abolish the auto-phosphorylation of CASK, presumably by reducing its ATP affinity (Fig. 4C). Inhibition by superstoichiometric Mg2+ suggests that the CASK CaM-kinase domain is optimally active in the presence of unchelated free ATP, and that the cellular Mg2+ content may regulate the kinase activity of CASK in vivo by limiting the free-ATP levels in the cytosol. To examine whether catalytic phosphotransfer can occur near neuronal levels of free Mg2+ (0.3–0.5 mM; (Brocard et al., 1993; Taylor et al., 1991)), we measured the autophosphorylation of the CASK CaM-kinase domain as a function of incrementally increased Mg2+-concentrations. We found that Mg2+ steeply inhibited the CaM-kinase domain autophosphorylation at low concentrations, but that the inhibition saturated at higher Mg2+-concentrations, such that a significant amount of residual autophosphorylation (>33% of maximum) remained even at high Mg2+ concentrations (Fig. 4D). The inhibition constant for Mg2+-dependent inhibition of phosphotransfer (KiMg2+ ≈ 61 μM; Fig. 4D) parallels that of the Mg2+-dependent inhibition of ATP binding (KiMg2+ ≈ 10.41 μM; Suppl. Fig. S6B), suggesting that the inhibition of phosphotransfer by Mg2+ is a result of the Mg2+-dependent disruption of ATP-coordination. These data indicate that the concentration of ATP-Mg2+ in the neuronal cytoplasm should allow physiological catalysis by CASK, but this catalytic activity of CASK may be regulated by changes in the cytosolic divalent ion concentration.
CASK is abundant in brain (Hata et al., 1996), and may associate with the pre-synaptic active zone (Olsen et al., 2006),. Since brain contains relatively high concentrations of free ATP (Foskett et al., 2007; Mak et al., 1999; Rubin and Brown, 1988; Taylor et al., 1991) and relatively low concentrations of free Mg2+ (Brocard et al., 1993; Gotoh et al., 1999; Kato et al., 1998), the Mg2+-dependent inhibition of the CASK CaM-kinase domain does not eliminate the possibility that this domain functions as a kinase. To test this, we examined the effect of CASK kinase activity on neurexin-1β (neurexin-1), a CASK-interactor at the synapses (Hata et al., 1996).
We first investigated the phosphorylation of the cytoplasmic C-terminal tail of neurexin-1 by the isolated CASK CaM-kinase domain (that is, the domain without the associated MAGUK domains) in vitro. Phosphorylation assays revealed that the CaM-kinase domain directly phosphorylates the neurexin-1 C-tail, and that this activity is strongly inhibited by Mg2+ (Suppl. Fig. S7). However, the phosphotransfer was minimal, and no phosphorylated peptide was detectable by mass spectrometry (data not shown).
Full-length CASK binds to neurexins via its PDZ-domain, and CASK co-localizes and co-precipitates with neurexins (Suppl. Fig. S8A,B and S9 (Hata et al., 1996)), suggesting that the interaction of CASK with neurexins via the PDZ domain may facilitate phosphotransfer. To test this hypothesis, we measured phosphotransfer to the neurexin-1 cytoplasmic tail in a complex with full-length CASK. Indeed, we found that neurexin-1 was efficiently phosphorylated (Fig. 5A). These experiments, performed with a vast excess of neurexin, nevertheless resulted in a phorphorylation stoichiometry of ≈ 17 %. Whereas, CASK autophosphorylation was undetectable in these reactions, possibly due to competition with a substrate. The net rate of phosphotransfer to neurexin-1 (Vmax ≈ 159.4 nmol/μmol enzyme/min) was 30-fold faster than the rate of CASK autophosphorylation, suggesting that neurexin, when complexed to CASK, is a substrate for the CaM-kinase domain (Figs. 5A,B). The KmATP of neurexin-1 phosphorylation by CASK (≈ 748.7 μM) was similar to that of autophosphorylation reactions, consistent with the notion that ATP utilization by CASK is not enhanced by the co-localization and availability of an optimal substrate (Fig. 5B). Improved efficiency of phosphotransfer to the CASK-associated neurexin-1 thus could be attributed to increased local substrate concentration mediated by the PDZ domain-mediated interaction. The high phosphorylation efficiency in CASK-neurexin-1 complexes allowed phosphopeptide mapping of the neurexin-1 C-tail by LC-MS-MS. Peptides comprising residues 422 to 433 of neurexin-1 (QPSSAKSANKNK) were detected with single and double-phosphates, establishing that the neurexin-1 C-tail is directly phosphorylated by CASK (not shown).
In CASK, the L27 domain links the CaM-kinase domain to the PDZ domain, but deletion of this linker region (CASK-linker) had no effect on neurexin phosphorylation (Suppl. Fig. S10). This result indicates that the L27 domain is not directly involved in CASK-mediated phosphorylation, although binding of Velis to this domain (Butz et al., 1998) may inhibit phosphorylation of a substrate bound to the PDZ-domain, an exciting possibility that would provide for additional regulation of CASK activity.
We next examined whether CASK can phosphorylate neurexin-1β under physiological conditions. Co-expression of a CASK-EGFP fusion protein with a neurexin-1β cherry fusion protein recruits CASK from the cytosol to the plasma membrane, and increases phosphorylation of neurexin-1β (Suppl. Fig. S8A,C). However, in these experiments we detected significant phosphorylation of neurexin-1β in the absence of CASK, suggesting that not surprisingly, other kinases also phosphorylate neurexin-1β.
We explored neurexin phosphorylation within a neuronal context in rat primary hippocampal cultures. Since β-neurexins are far more prevalent in these cultures, we decided to use these as our readout. Phosphorylated β-neurexins could be detected in mature neurons at 14 days in vitro (DIV) (Fig. 6A).
Cytosolic Mg2+ and Ca2+ concentrations are increased by synaptic activity (Brocard et al., 1993; Gotoh et al., 1999; Kato et al., 1998). β-neurexin phosphorylation increased dramatically (more than 2-fold) when synaptic activity-driven divalent ion fluxes were suppressed (with APV, a NMDA-receptor antagonist and TTX, a Na+-channel blocker) (Fig. 6A). These drugs had no effect on β-neurexin phosphorylation prior to the development of mature synapses (at 6 DIV) (data not shown). Since CASK, contrary to other kinases (Fig. 6A), is inhibited by divalent ions, enhanced neurexin phosphorylation upon synaptic inactivation is strongly suggestive of CASK kinase activity.
To test whether CASK is directly responsible for β-neurexin phosphorylation in neurons, we used a knockin-knockout approach. Primary hippocampal cultures from CASK knock-in mice, containing Lox-P sites flanking the CASK gene, were transduced with GFP-Cre recombinase fusion protein, using a lentiviral vector system (Atasoy et al., 2007). Compared to the neurons from littermate wild type mice (WT), β-neurexin phosphorylation was reduced by nearly 40 % in CASK-knockout (CASK KO) neurons (Fig. 6B).
Since CASK has been proposed to act as an adaptor for various molecules, including other kinases (Kaech et al., 1998; Lu et al., 2003), disrupted docking of another kinase in the CASK KO neurons could have resulted in the reduced phosphorylation of β-neurexins. To preserve the adaptor function of CASK, a dominant-negative approach was used – that specifically targeted the kinase activity of CASK CaM-kinase domain. We developed a catalytically impaired CASK-mutant that could act as a dominant-negative competitor, when introduced into a neuron. Most studies involving the inactivation of classical protein kinases target residues thought to be involved in either γ-phosphate or substrate positioning (for instance Asp of HRD motif). Since the γ-phosphate of ATP bound to CASK CaM-kinase domain is already disordered, likely due to the lack of Mg2+-coordinating Asp of DFG motif, such a strategy may be ineffectual (Supp Fig. S13). Thus, based on the stereochemistry of ATP-coordination within the nucleotide-binding pocket of CASK (Fig. 2B), we mutated Ser24 and Val26 to Asp and Leu, respectively (S24D, V26L). The CaM-kinase domain of CASK bearing this double mutation (referred to as CASKSV) behaved as a hypomorph in vitro, evident from the dramatic reduction in both, autophosphorylation (Suppl. Fig. S11A) as well as neurexin C-tail phosphorylation (Suppl. Fig. S11B). Importantly, CASKSV retained the ability to bind ATP (data not shown), suggesting that it adopts the active conformation, and presumably retains native molecular interactions. In particular, CASKSV still binds efficiently to neurexin-1 via its PDZ-domain (data not shown), thus acting as a dominant-negative competitor of the wild type kinase.
We transfected rat hippocampal neurons at high efficiency with wild-type CASK (CASKwt)or the CASKSV variant, as EGFP-fusion proteins (Fig. 6C). Expression of CASKwt produced an almost 2-fold increase in β-neurexin phosphorylation, whereas the expression of CASKSV caused an almost 2-fold decrease in β-neurexin phosphorylation (Fig. 6C). Since the expression of CASKwt and the kinase-hypomorph CASKSV assert diametrically opposite effects on the β-neurexin phosphorylation-state, this phosphotransfer is likely a direct consequence of CASK kinase activity. Together, these results demonstrate that CASK physiologically phosphorylates endogenous neurexins in neurons, and this phosphorylation is regulated by the synaptic acitivity-driven fluxes of divalent ions.
In this study, we describe the crystal structure and biochemical properties of the CaM-kinase domain of CASK. We demonstrate that CASK functions as an active protein kinase that phosphorylates neurexins – and presumably other target proteins – by an unusual and novel mechanism. These observations are not only important for our understanding of CASK, an enigmatic yet essential MAGUK protein with a CaM-kinase domain that is completely conserved in vertebrates, but also for our concept of “pseudokinases” in the kinome, which may turn out to be enzymatically active kinases with special properties.
Mg2+ acts as an obligate cofactor for ATP-binding and phosphotransfer in all known kinases (Adams, 2001; Waas et al., 2004). Here, we demonstrate that the CASK CaM-kinase domain is a physiologically active kinase that catalyzes phosphotransfer from ATP to proteins even in the complete absence of Mg2+. To our knowledge, CASK is the first kinase, indeed the first nucleotidase, known to perform catalytic phosphotransfer in the absence of Mg2+.
The structure of the CASK CaM-kinase domain determined here, and comparison of its structure with those of other kinases, illustrates that the CaM-kinase domain of CASK adopts a constitutively active conformation (Figs. 1 and and2,2, and Supp. Figs. S2 and S3). Biochemical and enzymatic assays demonstrated that CASK binds ATP independent of Mg2+, and catalyzes autophosphorylation and neurexin-1 phosphorylation in the absence of Mg2+ (Figs. 3, ,44 and and5).5). Compared to other kinases, CASK contains non-canonical residues at several conserved positions of the nucleotide-binding pocket, which may account for its unusual catalytic mechanism. Both of the classical metal-coordinating residues are substituted (Asn146Cys and Asp162Gly; see Supp. Fig. S1). Moreover, Glu143 of the catalytic-loop directly coordinates the metal ion in DAPK1; while in CASK, this Glu is altered to His (Glu145His), which might repel Mg2+ electrostatically. These changes likely contribute to the absence of Mg2+ coordination in the CASK CaM-kinase domain, resulting in the de-coordination of β- and γ-phosphates of ATP in the presence of Mg2+. However, in the nucleotide binding pocket of CaMKII, the adenine base of ATP makes the most important contacts for the positioning of nucleotide (Kwiatkowski and King, 1987). A similar observation was made for the binding of 3'-AMP in one of our CASK CaM-kinase structures (Supp. Fig. S3). Thus, an altered Mg2+-coordinating sequence does not exclude ATP-binding and, as shown here, does not exclude catalysis. Importantly, similar to the CASK CaM-kinase domain, other “pseudokinase” domains with non-canonical Mg2+-binding motifs may coordinate ATP and phosphorylate physiological substrates as well.
CASK is also unusual compared to any other member of the CaM-kinase family in that its CaM-kinase domain exhibits a constitutively active conformation (Supp. Fig. S2). In an archetypal CaM-dependent kinase, the catalytic domain is followed by an autoinhibitory domain that constitutively inhibits kinase activity, but is disinhibited by Ca2+/calmodulin binding (Goldberg et al., 1996). Indeed, the CASK CaM-kinase domain is followed by a sequence that is homologous to the autoinhibitory domain of CaM-kinases (Supp. Fig. S1), which also binds Ca2+/calmodulin (Hata et al., 1996). However, unlike typical CaM-kinases, the autoinhibitory helix (αR1) of CASK does not engage in direct contacts with the ATP-binding cleft (Fig. 1 and Supp. Fig. S3). We also did not discern evidence for further C-terminal residues interacting with the ATP binding cleft, as in CaMKI, and we detected no stimulatory effect of Ca2+ and/or calmodulin on the CASK kinase activity (data not shown). In CaM-kinases, the autoinhibitory domain acts by competition with the substrate. The pseudosubstrate region of the CASK sequence that is homologous to the autoinhibitory domain contains Arg288Leu and Gln289Pro substitutions that result in a sub-optimal pseudosubstrate helix (Supp. Fig. S1; (Smith et al., 1992)) and is missing part of the consensus substrate motif (RXXT/S). Thus, the CASK CaM-kinase domain appears to retain a non-functional autoinhibitory domain as an evolutionary vestige of CaM-kinases. CASK, therefore, differs from other, evolutionarily closely related CaM-kinase domains not only in its Mg2+-independence, but also in its inherently “closed” active conformation which binds to nucleotides constitutively.
An almost essential consequence of the constitutively active conformation of the CASK CaM-kinase domain is that the domain exhibits a very low catalytic rate, as shown in autophosphorylation measurements (Fig. 4) and in measurements of neurexin-1 phosphorylation by the isolated CaM-kinase domain of CASK lacking its neurexin-binding PDZ-domain (Suppl. Fig. S7). Mechanistically, this low rate is likely due to the loss of Mg2+-coordination by the domain. The low catalytic rate of the CASK CaM-kinase domain presumably serves to ensure that the kinase does not phosphorylate potential substrates randomly. We show that the phosphorylation rate of neurexin-1 is increased dramatically, however, when full-length CASK forms a complex with neurexin-1 via the PDZ-domain of CASK (Fig. 5). The binding of neurexin-1 to CASK probably increases the phosphotransfer efficiency by increasing the substrate specificity and the local substrate concentration. This result suggests a general mechanism for CASK kinase activity, whereby CASK couples an intrinsically slow but constitutively active kinase domain to a PDZ-domain that recruits the substrates to the kinase domain, thereby increasing the local substrate concentration by several orders of magnitude (see model in Fig. 7). According to this model, CASK unites two separate functions – the recruitment activity of MAGUKs and the kinase activity of the CaM-kinase domain – into a single unit whose objective is phosphorylation of specific interacting proteins (Fig. 7).
CASK phosphorylates neurexin-1 in vitro and in vivo in a reaction that depends on a catalytically active CaM-kinase domain (Fig. 5 and and6).6). β-neurexins, described here as a substrate of CASK, is a pre-synaptic cell-adhesion molecule ((Nam and Chen, 2005; Ushkaryov et al., 1992); reviewed in (Missler et al., 2003)). Its heterotypic binding to the post-synaptic neuroligins may be involved in synaptic function (Missler et al., 2003), and could induce synapse formation even on non-neuronal co-cultured cells (Graf et al., 2004). The neurexin-neuroligin interaction is a candidate for synaptic specialization and pre-post synapse communication (reviewed in (Cline, 2005; Lise and El-Husseini, 2006; Yamagata et al., 2003)). Both neurexin (via linkage and copy number analyses) and neuroligin mutations have been linked to autism spectrum disorders in human cohort studies (Jamain et al., 2003; Szatmari et al., 2007). Deletion of CASK may be connected to X-linked optic atrophy and mental retardation, according to human mapping and cohort studies (Dimitratos et al., 1998; Froyen et al., 2007). The evolutionary conservation of CASK and neurexins, and their central importance for survival and synaptic function in mice (Atasoy et al., 2007; Missler et al., 2003) indicates that neurexin phosphorylation by CASK may be crucial to neuronal function.
In addition to the control of CASK kinase activity by the PDZ-domain mediated substrate recruitment, we examined whether it is regulated by synaptic activity-driven rise in Ca2+ and Mg2+ levels (Brocard et al., 1993; Gotoh et al., 1999; Kato et al., 1998). In neurons, application of glycine with glutamate augments the free-Mg2+ level from a resting 0.5 mM up to 11 mM – a 22-fold increase (Brocard et al., 1993); in addition to the activity-driven Ca2+ influx. Cytosolic free-ATP levels, and thus the CASK kinase activity, could be regulated by these changes in divalent ions (Kargacin and Kargacin, 1997). Only in synaptically active mature neurons, we observed a strong increase in neurexin phosphorylation upon silencing synapses, indicating that contrary to other kinases, CASK kinase is inhibited by neuronal activity (Fig. 6A). We envision that CASK kinase activity is maximal during neuronal development and synaptogenesis, and declines with the onset of synaptic function, but is reactivated when neurons are silent. This developmentally regulated activity is in line with the phenotypic defects in CASK knockout mice (Atasoy et al., 2007) as well as the developmental nature of CASK- and neurexin-related pathologies (Froyen et al., 2007; Szatmari et al., 2007).
Moreover, CASK is expressed ubiquitously at low levels (Hata et al., 1996). The non-neuronal functions of CASK are evident from developmental defects in CASK/Lin-2-null animals, such as cleft palate in mice (Atasoy et al., 2007) and vulval dysgenesis in C. elegans (Hoskins et al., 1996). The rescue experiments in C.elegans revealed that point mutants of the CASK CaM-kinase domain were significantly less capable of rescuing the Lin-2 phenotype despite 2–5 fold overexpression, whereas mutations in the guanyl kinase domain had no effect on the rescue (Hoskins et al., 1996). Thus, the developmental defects that arise in the absence of CASK might be linked to the kinase function of CASK CaM-kinase domain. In non-neuronal cells, CASK-interacting adhesion molecules of the syndecan or JAM families could be substrates (Cohen et al., 1998; Hsueh et al., 1998; Martinez-Estrada et al., 2001). These molecules share the PDZ-domain mediated CASK-association, and at least in the case of syndecan-2, serine residues in the cytoplasmic tail homologous to those of neurexins are phosphorylated in vivo (Itano et al., 1996).
Kinase activity of CASK suggests the possibility of regulated phosphorylation of other interactors as well, such as Mint-1, liprin, Ether-a-gogo, PARKIN and CaMKII (Fallon et al., 2002; Hodge et al., 2006; Lu et al., 2003; Marble et al., 2005).
Finally, of the 518 known kinases in the human genome, 48 are predicted to be pseudokinases (Boudeau et al., 2006). Each of these have one or more of the invariant motifs altered through evolution. Nine of the presumed pseudokinases, including CASK, lack a canonical DFG motif. Furthermore, this motif is altered along with other canonical motifs (HRD and/or VAIK) in 22 additional pseudokinases. Therefore, other pseudokinases, especially those with atypical DFG motifs, could be active in physiologically relevant environments; suggesting that the catalytically active kinome may be more diverse than originally envisioned.
Experimental Procedures for structural and mass-spectrometric studies can be found in the Supplementary Material.
TNP-ATP was acquired from Molecular Probes Inc. (Eugene, OR, USA). Experiments were performed in 50 mM Tris-HCl, pH 7.2, 50 mM KCl in 1 cm × 1 cm fluorescence cuvettes at 25°C using a Jobin Yvon-Spex Fluoromax-2 (Stewart et al., 1998). Samples were excited at 410 nm, and emission spectra scanned from 500–600 nm. Excitation and emission slits were set at 3 and 5 nm, respectively. For TNP-ATP titration experiments, fluorescence emission at 541 nm was measured and corrected by subtracting the signal from a TNP-ATP buffer control, in which BSA replaced CASK CaM-kinase domain. For magnesium titration experiments 1 μM of TNP-ATP was mixed with 1 μM protein in a cuvette and emission was recorded at 541 nm after addition of defined quantities of Mg2+.
In the Tris-Cl buffer supplemented with 2 mM EDTA, recombinant CASK CaM-kinase domain was incubated for increasing lengths of time with 1 mM γ32P-ATP (2 × 107 cpm) for 1 h at RT. For determining the KmATP increasing concentration of ATP was added for 30 min. The protein was separated with SDS-PAGE and visualized on a phosphorimager. The radioactivity incorporated in the band was determined for quantification. Coomassie staining was used for loading control.
GST-neurexin-1 C-tail, which comprised the entire cytoplasmic domain of neurexin-1 (AMYKYRNRDEGSYHVDESRNYISNSAQSNGAVVKEKQPSSAKSANKNKKNKDKEYYV), was bound to glutathione beads and incubated overnight with full-length CASK, in Tris-Cl buffer at 4 °C. After washing 3 times in ice-cold Tris-Cl buffer, the complexes were incubated with indicated concentrations of γ32P-ATP in Tris-Cl buffer supplemented with EDTA (2 mM) or Mg (2 mM) at RT for indicated times. The proteins were eluted from the beads by thrombin cleavage, separated with SDS-PAGE and visualized on a phosphorimager. Radioactivity incorporated was calculated by correlating the beta counts in a band (measured with a Beckman-Coulter LS-6000 scintillation system) with its phosphorimager quantification (performed with Molecular Dynamics Storm scanner and Image-Quant software).
HEK293T cells were co-transfected with EGFP-CASK in pEGFP-C3 vector and neurexin-1β-cherry in pCAAG vector or the empty pCAAG vector using FUGENE-6 (Roche). 2 days post-transfection, the cells were washed in PBS and fixed in 4 % paraformaldehyde in PBS at RT for 20 min. Laser scanning confocal microscopy was performed to compare localization, on a Leica TCS SP-2, inverted microscope.
Primary mixed hippocampal cultures from P1 rat pups were transfected on 6 days in vitro (DIV) using a modified Ca2+-phosphate method. Two days later, the cells were kept in 500 nM TTX and 50 μM APV overnight. Subsequently, the cells were washed and incubated in phosphate-free depletion-buffer for 30 min at 37 °C (10 mM HEPES-NaOH, pH 7.2, 150 mM NaCl, 4 mM KCl, 2 mM MgCl2, 2 mM CaCl2, 10 mM D-glucose, 100 μM insulin), followed by incubation in the same buffer, supplemented with 100 μCi 32P-orthophosphate for 1 h. Cells were washed twice with phosphate-free buffer and lysed in ice cold solubilization buffer (10 mM Tris-Cl, pH 6.8, 150 mM NaCl, 1 % Triton-X100, 4 mM EDTA), supplemented with protease inhibitor cocktail (leupeptin, pepstatin, PMSF and aprotinin), phosphatase inhibitor cocktails-1 and 2 (Sigma). Debris was spun down (14,000 rpm for 10 min at 4 °C) and supernatant was incubated with immobilized GST-CASK PDZ-domain overnight. Complexes were washed in solubilization buffer 3 times and separated by SDS-PAGE, followed by phosphorimager scanning. β-neurexins immunoblots (pan- β-neurexins rabbit polyclonal antibody (Suppl. Fig. S12) were used as loading control. Neuronal lysates were also separated and immunoblotted for CASK with mouse monoclonal antibody (Transduction Labs).
Conditional CASK knockin-knockout mice were generated and used for primary hippocampal neuronal culture as described (Atasoy et al., 2007). In brief, primary hippocampal cultures from CASK knock-in mice, containing Lox-P sites flanking the CASK gene, were transduced with GFP-Cre recombinase fusion protein, using a lentiviral vector system at DIV 4. Six days later (DIV 10), in vivo 32P-phosphorylation assay was performed as described above.
HEK293T cells were co-transfected with EGFP-CASK and flag-tagged neurexin-1β. Phosphorylation and lysis were performed as above, with the exception of TTX and APV treatment. Neurexin-1β was batch precipitated on anti-flag (M-2) beads (Sigma), separated on SDS-PAGE and quantified by phosphorimager. Coomassie staining was used for visualizing total proteins loaded.
We thank Drs. Elizabeth Goldsmith, A. J. Robison, Mark Etherton, Gottfried Mieskes, Antony Boucard and Ralf Jauch for comments on the manuscript; members of the Südhof and Jahn laboratories for discussions; Lin Fan, Andrea Roth and Izabella Kornblum for technical support; and Reinhard Lührmann for the use of crystallography equipment. This work was supported by a grant from the NIMH (R37 MH52804-08 to TCS) and the Max-Planck-Society. MS is a Human Frontiers Long Term Fellow.
Supplementary Material Supplementary Materials include one table, thirteen figures, and Experimental Procedures for structural and mass-spectrometric studies.