A variety of viral and signal transduction proteins are known to be myristoylated. Although the role of myristoylation in protein–lipid interaction is well established, the involvement of myristoylation in protein–protein interactions is less well understood. CAP‐23/NAP‐22 is a brain‐specific protein kinase C substrate protein that is involved in axon regeneration. Although the protein lacks any canonical calmodulin (CaM)‐binding domain, it binds CaM with high affinity. The binding of CAP‐23/NAP‐22 to CaM is myristoylation dependent and the N‐terminal myristoyl group is directly involved in the protein–protein interaction. Here we show the crystal structure of Ca2+‐CaM bound to a myristoylated peptide corresponding to the N‐terminal domain of CAP‐23/NAP‐22. The myristoyl moiety of the peptide goes through a hydrophobic tunnel created by the hydrophobic pockets in the N‐ and C‐terminal domains of CaM. In addition to the myristoyl group, several amino‐acid residues in the peptide are important for CaM binding. This is a novel mode of binding and is very different from the mechanism of binding in other CaM–target complexes.
Since the discovery of protein myristoylation in the catalytic subunit of cAMP‐dependent protein kinase (Carr et al, 1982), various viral and signal transduction proteins have been shown to be fatty acylated (Johnson et al, 1994). This modification is often essential for the proper functioning of the modified protein. For example, the transforming activity of p60src from Rous sarcoma virus is dependent on its myristoylation (Garber and Hanafusa, 1987). However, the mechanism by which this modification exerts its effects is still largely unknown (Johnson et al, 1994; Resh, 1996, 1999). It is generally assumed that hydrophobic acyl groups, such as myristoyl and palmitoyl groups, are often involved in protein–membrane interactions. Due to its intermediate hydrophobicity, myristoylation has been implicated in reversible membrane association of proteins (McLaughlin and Aderem, 1995; Taniguchi, 1999). Studies from our own and others' laboratories have established that such a mechanism, in fact, participates in the phosphorylation‐dependent interaction of myristoylated alanine‐rich C kinase substrate (MARCKS) with membranes (Taniguchi and Manenti, 1993; Kim et al, 1994). In the case of recoverin, the binding of Ca2+ induces a drastic conformational change of the protein, causing the myristoyl group hidden inside the protein to protrude, whereupon it can interact with the membrane (Ames et al, 1997). The modification has also been shown to affect the stability of cAMP‐dependent protein kinase (Yonemoto et al, 1993). On the other hand, how protein myristoylation is involved in protein–protein interaction remains unclear, even though it has been extensively studied.
Brain‐specific protein CAP‐23/NAP‐22 was characterized first in chicken brain as a cortical cytoskeleton‐associated protein of 23 kDa (Widmer and Caroni, 1990). Its rat homologue was later identified as a neuron‐specific acidic protein of 22 kDa (Maekawa et al, 1993). Although the physiological function of the protein has not yet been determined, it is known to be involved in spinal axon regeneration and is found in the brain membrane raft fraction (Maekawa et al, 1999; Laux et al, 2000; Bomze et al, 2001). Like the two related proteins GAP‐43 and MARCKS, CAP‐23/NAP‐22 is a prominent substrate of protein kinase C (PKC), and its phosphorylation by PKC is regulated by calmodulin (CaM) binding (Maekawa et al, 1994). Using synthetic myristoylated peptides, as well as myristoylated and nonmyristoylated recombinant proteins, we narrowed down the CaM‐binding domain of CAP‐23/NAP‐22 to the myristoylated nine‐amino‐acid N‐terminal peptide (Takasaki et al, 1999). Interestingly, the binding of CAP‐23/NAP‐22 to CaM is dependent on myristoylation; only the myristoylated recombinant protein binds to CaM. Thus, this is the first definitive demonstration that protein myristoylation is involved in protein–protein interaction.
Here we show the crystal structure of Ca2+‐CaM bound to a myristoylated peptide derived from the N‐terminal domain of CAP‐23/NAP‐22. The obtained structure revealed that the myristoyl moiety of CAP‐23/NAP‐22 enters into a hydrophobic tunnel created by the hydrophobic pockets of CaM. In addition to the myristoyl group, several amino‐acid residues in the peptides are important in the binding of CaM. The overall binding mode is distinct from those observed in other protein–protein complexes. These results reveal a unique structure and demonstrate how protein myristoylation functions in protein–protein interaction.
Results and discussion
Overall structure of the myristoylated CAP‐23/NAP‐22 peptide–CaM complex
Our previous work demonstrated that the myristoylated nine‐amino‐acid N‐terminal peptide of CAP‐23/NAP‐22 is sufficient for high‐affinity binding to Ca2+‐CaM (3.0 nM) and, moreover, that the myristoyl moiety of CAP‐23/NAP‐22 is required for binding (Takasaki et al, 1999). To understand the structural basis of the myristoylation‐dependent binding, we attempted to crystallize the complex of CaM and the myristoylated CAP‐23/NAP‐22 peptide (myr‐G201GKLSKKKK209, mNAP peptide, first N‐terminal Gly defined as G201). We successfully crystallized the CaM–mNAP complex and solved the X‐ray structure with a 2.3 Å resolution (Figure 1, Table I). Besides a few surface side chains and the residues of the central linker regions, the structure is well defined by the electron density. The overall structure of CaM in the complex is an ellipsoidal compact structure with a cavity, mainly filled with the myristoyl moiety of the mNAP peptide. The conformation of CaM is similar to those in other CaM–peptide or CaM–drug complexes, with the two lobes of CaM folded in such a way that they can grab the target peptide (Ikura et al, 1992; Meador et al, 1992, 1993; Vandonselaar et al, 1994; Osawa et al, 1999; Harmat et al, 2000).
Mechanism of the binding mode of the myristoyl moiety
The structure of the myristoylated peptide is exceptional, as expected. In the complex, the peptide consists of a long myristoylated moiety, a hairpin‐like loop (residues 201, 202) and an extended region (residues 203–206) (Figure 1, see also Figure 3). The myristoyl moiety of the mNAP peptide is unambiguously defined by electron density (Figure 2A). While the extended region travels along the groove between the two lobes of CaM, the myristoyl moiety goes through the hydrophobic tunnel created by the hydrophobic pockets of the N‐ and C‐terminal lobes of CaM (Figure 2B). Residues of CaM within 5 Å of the myristoyl group are mainly hydrophobic (Figure 2C). In particular, CaM residues Leu39 and Phe92 contribute to binding of the head (C10–C13) of the myristoyl moiety. Residues Phe19 and Leu112 interact with the middle part (C7–C9), while residues Leu18 and Met109 contribute to binding of the neck (C4–C6) of the myristoyl moiety. The myristoyl group is thus firmly anchored to CaM by multiple hydrophobic interactions.
Until now, few three‐dimensional structures of myristoylated proteins have been reported. The NMR structure of recoverin in a Ca2+‐free state revealed the direct interaction of the acyl group with its own protein moiety (Tanaka et al, 1995). In addition, the crystal structure of myristoylated cAMP‐dependent protein kinase has elucidated the structural basis for the effects of acylation on protein stability (Zheng et al, 1993). In these structures, the myristoyl moieties are sequestered deeply into hydrophobic pockets formed by many hydrophobic residues. However, these are intramolecular interactions wherein the myristoylated protein interacts with its own acyl group. Our structure is unique because it shows how a protein grabs the covalently bound acyl chain of another protein. Similar myristoylation‐dependent protein–protein interactions are thought to play roles in the structural organization or stability of picornavirus capsid protein VP4 (Chow et al, 1987) and calcineurin B (Griffith et al, 1995). The hydrophobic pocket observed in our structure is reminiscent of that in the isoprenylated Rho‐GDI‐binding pocket of cdc42 (Gosser et al, 1997). However, the myristoyl‐binding pocket in our structure is much larger and contains many more hydrophobic residues, suggesting that hydrophobic interactions are highly important for myristoylation‐dependent protein–protein binding.
Interaction of peptide amino acids with CaM
We have previously shown that neither myristic acid nor a short myristoylated peptide, myr‐GGK, binds to CaM (Takasaki et al, 1999). This suggests that the binding of the myristoylated N‐terminal peptide is affected not only by a simple hydrophobic interaction between the myristoyl group and the hydrophobic part of CaM but also by other specific interactions. The present structure provides an insight into the structural basis of the interaction between the N‐terminal peptide of NAP‐22 and CaM (Figure 3A). The two N‐terminal Gly residues of the mNAP peptide (Gly201 and Gly202) form a tight hairpin turn, which resembles a hook, positioning the following two amino acids, Lys203 and Leu204, inside of the CaM cavity. Replacing one of the Gly residues with other amino acids greatly reduced the affinity (data not shown), indicating that the presence of two successive Gly residues is essential for binding. The presence of other amino acids makes such a turn energetically unfavorable. The only hydrophobic residue in the N‐terminal peptide, Leu204, interacts with the side chains of Phe12 and Met144 of CaM through hydrophobic interactions. Furthermore, it appears that there is an intermolecular electrostatic interaction between the basic residue, Lys203, of the peptide and the negatively charged CaM residue, Glu7. Thus, in addition to the myristoyl group, several N‐terminal residues play important roles in the NAP‐22–CaM interaction. Unfortunately, no significant electron density is observed for the three C‐terminal Lys residues of the peptide. However, many acidic residues of CaM surround the C‐terminal basic cluster of the mNAP peptide (Figure 3B), indicating that electrostatic interactions can provide additional binding energy. In fact, successive removal of these Lys residues gradually reduces the affinity of the mNAP peptide for CaM (Takasaki et al, 1999).
Novel mode of target recognition by CaM
The majority of the CaM‐binding domains contain positively charged and hydrophobic residues, and many of them have the potential to form an amphiphilic α‐helix (Crivici and Ikura, 1995; James et al, 1995). Indeed, previously determined structures show that CaM‐bound target peptides form α‐helices in the CaM complex. However, in the present structure, the peptide moiety of the mNAP peptide does not form an α‐helix in the complex (Figure 1). The present structure was compared with that of a typical target peptide–CaM complex (CaM kinase II peptide–CaM complex) (Figure 4A). In the comparison, the N‐terminal domains of CaM were aligned to minimize the differences. Interestingly, the myristoyl moiety of the mNAP peptide almost coincides with the α‐helical peptide of CaM kinase II (Figure 4B). However, there are significant differences between them, especially the position of hydrophobic residues involved in the interactions. In the canonical CaM‐binding domains, two key hydrophobic residues are positioned 9, 13 or 15 residues apart. These two residues are anchored to the hydrophobic pockets formed in each of the two CaM lobes (Figure 4C). In the hydrophobic pockets of the CaM C‐terminal lobe, specific hydrophobic residues of the target peptides can be seen (Leu299 of the CaMKII peptide, Trp800 of the MLCK peptide and Phe157 of the MARCKS peptide), which are located in the same space (Figure 5A). Although our recent structure of the MARCKS peptide–CaM complex shows a novel CaM‐binding mode, this hydrophobic pocket of the C‐terminal lobe is present (Yamauchi et al, 2003). Furthermore, similar hydrophobic pockets in the N‐terminal lobe of CaM are formed by the key hydrophobic residues (Leu308 of the CaMKII peptide and Leu813 of the MLCK peptide) (Figure 5B). In contrast, the two key hydrophobic residues in the mNAP peptide–CaM complex are absent in both N‐ and C‐terminal lobes of CaM (Figure 5A and B). Thus, our present structure is distinct from the hydrophobic pockets formed in other CaM complexes because the myristoyl moiety and Leu204 take the place of the two hydrophobic residues. Furthermore, the parts of CaM involved in the interaction are quite different than in other complexes, because many hydrophobic residues are aligned to form a single large hydrophobic cavity to accommodate the more bulky myristoyl moiety.
The recently reported structures of three CaM–target complexes, Ca2+‐activated K+ channel (SK2)–CaM, edema factor (EF)–CaM and plant glutamate decarboxylase (GAD)‐CaM, show that CaM can take completely different, nonglobular conformations (Schumacher et al, 2001; Drum et al, 2002; Yap et al, 2003). In these structures, there are unique domain orientations of CaM. This is especially evident when comparing the angles between the helices flanking the linker (Yap et al, 2003). The interhelical angles for the classical CaM–peptide complexes range between −114.2 and −123.2°, while the values for SK2, EF and GAD complexes are 78.0, 47.8 and −26.3°, respectively (Yap et al, 2003; Table II). However, the interhelical angle for the mNAP peptide‐CaM is −110.1°, suggesting that the domain orientation resembles that of classical CaM–peptide complexes. Furthermore, the distance between residues 72 and 85 for the mNAP peptide–CaM complex (19.3 Å) is almost identical to that for classical CaM–peptide complexes (Table II). Thus, the present structure shows that the globular form of Ca2+‐CaM can be generated when it is complexed with the mNAP peptide, because the myristoylated target uses a broad hydrophobic interaction instead of the conventional hydrophobic pocket. Overall, the CaM–mNAP peptide complex represents a novel mode of target recognition for CaM and demonstrates the surprising versatility of this important signaling molecule.
Implications for myristoylation‐dependent interaction
CAP‐23/NAP‐22 may not be the only example where a myristoyl group is involved in a protein–protein interaction. Recently, we demonstrated that the N‐terminal regions of the HIV‐1 Nef protein and MARCKS bind to CaM (Hayashi et al, 2002; Matsubara et al, 2003). The N‐terminal regions of these proteins share structural features, including one hydrophobic residue surrounded by a few basic residues that are preceded by the myristoyl group. This raises the possibility that the protein‐bound myristoyl group is directly involved in protein–protein interactions. Recent biochemical evidence has shown that CAP‐23/NAP‐22 and MARCKS, which share many structural features, accumulate in a neuronal membrane raft fraction (Maekawa et al, 1999; Laux et al, 2000; Khan et al, 2003). In the membrane rafts, these proteins are regulated by CaM binding and PKC‐mediated phosphorylation in the N‐terminal domain. In the case of HIV‐1 Nef, the protein also exists in the membrane rafts after infection by the HIV‐1 virus (Wang et al, 2000; Zheng et al, 2001). The myristoylation may help recruit these proteins to the membrane raft structures by means of both protein–protein and protein–lipid interactions.
Owing to the hydrophobicity and the structural resemblance to the lipids, the roles of fatty acylation in the protein–lipid interaction are easy to understand. The roles of fatty acylation in protein–protein interactions, on the other hand, have remained controversial. Our previous biochemical and present structural studies clearly establish the direct involvement of fatty acylation in protein–protein interaction. The present structure of the CaM–mNAP peptide complex explains how protein myristoylation mediates protein–protein interaction. Comparison with other structures such as recoverin and cAMP‐dependent protein kinase, in which proteins bind to their own myristoyl groups, should provide a framework for understanding the molecular basis of lipid modification recognition by proteins.
We recently reported the crystal structure of the MARCKS effector domain complexed with CaM (Yamauchi et al, 2003). MARCKS and CAP‐23/NAP‐22 belong to the same protein family of acylated hydrophilic and acidic proteins (McLaughlin and Aderem, 1995; Taniguchi, 1999). In contrast to CAP‐23/NAP‐22, MARCKS has a domain in the middle of its structure that binds various cellular components, including CaM and membrane phospholipids. The structure of a peptide derived from this domain complexed with CaM was novel, because it does not form a conventional α‐helix (Matsubara et al, 1998; Yamauchi et al, 2003). The structure of the MARCKS peptide/CaM complex clearly lacks one of the two hydrophobic pockets (Yamauchi et al, 2003). This and our present work illustrate the versatility of CaM in recognizing a wide variety of target structures. This degree of structural versatility is beyond that previously predicted for CaM (Crivici and Ikura, 1995; James et al, 1995). Myristoylated proteins have diverse functions and intracellular destinations, and these are clearly regulated by this modification due to alterations in both protein–lipid and protein–protein interactions. Finally, our structure may help in understanding intracellular signaling and pathological processes, because protein myristoylation not only regulates the function and localization of signaling molecules but also has been implicated in cancer, viral disorders and infections. In this way, these studies may assist in the future development of therapeutic agents for cancer and infectious diseases.
Materials and methods
Crystallization, data collection and structure determination
A myristoylated peptide of nine amino acids, based on the N‐terminal sequence of CAP‐23/NAP‐22 (myr‐GGKLSKKKK) and synthesized using standard Fmoc chemistry, was purchased from Research Genetics. Recombinant human CaM was expressed in Escherichia coli and purified to homogeneity by column chromatography on phenyl‐sepharose. CaM was dialyzed against H2O and lyophilized. For crystallization (hanging‐drop vapor diffusion), 0.9 mM CaM and 2.7 mM peptide in 10 mM CaCl2 were mixed 1:1 with reservoir solution (36% PEG1000, 20 mM cacodylate buffer, pH 5.5) and equilibrated against 1 ml of reservoir solution at 4°C. The crystal growth took 2–3 weeks. Glycerol (20%) was used as the cryoprotectant. Reflections from the structure were collected from a single crystal at 90 K using an image plate detector and a SPring‐8 BL45XU beamline. Data processing was carried out using DENZO and Scalepack (Otwinowsky and Minor, 1997). The space group is P3221 with cell constants a=b=40.6 Å and c=173.6 Å, wherein there is one CaM–peptide complex in the asymmetric unit. The structure was determined by molecular replacement using the AMoRe program of the CCP4 package (CCP4, 1994) and the protein portion of the CaM–arylalkylamine complex as a model for the CaM structure (Harmat et al, 2000).
The model‐building steps were carried out with the TURBO‐FRODO (Roussel and Cambillau, 1991) program and the resulting model was refined with CNS (Brunger et al, 1998). We used rigid body refinement followed by torsion‐angle‐simulated annealing at a starting temperature of 4000 K (slow cooling) to 2.3 Å. The progress of model building and refinement was monitored using Rfree. After several rounds of model building and refinement, the refinement converged to R=24.7% and Rfree=27.4%. The final structure consists of 144 CaM residues, four Ca2+ ions and one myristoylated peptide, with an RMS deviation of 0.006 Å in bond length and 1.07° in bond angles. PROCHECK showed that the final model lacked residues in disallowed regions. Interhelical angles and distances were calculated by VGM (Yap et al, 1999, 2002).
PDB accession code
The coordinates and structure factors have been deposited in the Protein Data Bank under ID code 1L7Z.pdb.
This work was supported by Grants‐in‐Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan.
- Copyright © 2004 European Molecular Biology Organization