We have determined the first crystal structure of human CRMP-2, by far the best characterised member of the CRMP family, with roles in neurological development and disease. CRMP-2 is a homotetramer with high structural homology to CRMP-1 and the DHPs. Based on the analysis of the tetramer interfaces in the crystal structures of CRMP-1 and CRMP-2 homotetramers, we are able to propose a model for CRMP-1/2 heterotetramers, where two CRMP-1 and two CRMP-2 subunits arrange into homodimers via interface 2. These homodimers then form a heterotetramer via interface 1. The model for the heterotetramer is presented in Fig. 6.
As is the case with CRMP-1 (Deo et al. 2004), the tetramerisation of CRMP-2 is similar to that of DHP, apart from minor conformational differences e.g. in the small domain and the presence of the C-terminal helix that is also involved in oligomerisation. This C-terminal helix is the only region on the CRMP-2 surface with a positive charge potential (Fig. 1b). Regarding DHP tetramerisation, it is important to note that in fact, the human DHP sequence is highly homologous to that of CRMP-1 and CRMP-2 (Fig. 3c) also in the region of the C-terminal helix, and residues involved in tetramerisation in this region are conserved. Thus, human DHP most likely also has a C-terminal helix such as that seen in the CRMPs, and fine details of tetramerisation probably more closely resemble those of the CRMPs than those of previously determined DHP structures from yeasts and bacteria. CRMP-2 is also able to interact directly with DHP (Wang and Strittmatter 1997), and it is likely that CRMP-2 is able to regulate DHP activity via hetero-oligomer formation with active DHP.
The pocket corresponding to the active site of DHP is also present in CRMP-2, within the TIM barrel. No small molecule ligands for CRMP-2 have been reported, but this pocket, which is rather different in CRMP-1 and CRMP-2, could be used to bind such ligands. Whether calcium, which was required to obtain CRMP-2 crystals (data not shown) and was found in this pocket, could be a physiological ligand for CRMP-2, needs to be determined. It is likely that the binding is non-specific, taking into account both the high experimental Ca2+ concentration and the fact that only a single calcium-coordinating oxygen comes from the CRMP-2 protein. We plan to further study the ligand-binding properties of CRMP-2, with a specific emphasis on this pocket.
N- and C-terminal domains of CRMP-2
The mutation of the loop on the top of the small domain (residues 46–57) to polyalanine in CRMP-1 was shown to generate a constitutively active form of the protein (Deo et al. 2004). The sequence of this region is identical in CRMP-2. Comparing the loop in CRMP-1 and CRMP-2 indicates a slightly different conformation, and overall, it can be concluded that the loop is not tightly fixed onto the top of the small domain; for example, Leu49 has limited hydrophobic interactions despite its pointing into the protein (Fig. 7a). In addition, in our structure, all side chains of the loop were defined in electron density and could be modelled, leading to a more accurate model of this functionally important region. In the homologous urease α subunit, this loop is much longer and interacts closely with the β subunit (Benini et al. 2001). Thus, this loop may be a conserved site for protein interactions in this structural family.
Figure 7. Previously mapped interaction domains of collapsin response mediator protein 2 (CRMP-2) in the structure. (a) The loop (residues 46–57) in the small N-terminal domain is well defined in the electron density. Cyan-CRMP-2, grey-CRMP-1. The electron density is that of the final 2Fo–Fc map, contoured a 1 σ. For CRMP-1, most side chains are not present in the model, and for example Leu49 was modelled to point outwards from the loop. In CRMP-2, Leu49 is well-defined, together with the rest of the loop. (b) Overall view with coloured mapped domains for protein interactions, such as determined mainly by using truncation mutants in previous studies. Note how the previously suggested minimal binding sites for both Numb (275–322, blue) and tubulin (323–381, red) are buried within the CRMP-2 fold. The loop on top of the small domain (residues 46–57) is shown in orange.
Download figure to PowerPoint
Alternative splicing has been reported to generate two isoforms of CRMPs 1–4, differing in their N-terminal sequences (Yuasa-Kawada et al. 2003). The structure presented here corresponds to the shorter, better characterised, isoform, occasionally called CRMP-2B in the literature. Our results, thus, confirm that the folded region which is comprised of the small domain and the TIM barrel domain (residues 1–489) is enough for tetramer formation, and thus, neither the extended N-terminus for the alternatively spliced forms nor the unfolded C-terminal region is required for oligomerisation. The extra sequence present in the longer form, CRMP-2A, bears no significant similarity to proteins outside the CRMP family, apart from a slightly extended homology to the urease α subunit (data not shown). Interestingly, the two splice isoforms of CRMP-2 have opposite effects on microtubule organisation and axonal growth, indicating that the N-terminal extension is functionally important in regulating CRMP-2 activity (Yuasa-Kawada et al. 2003). Furthermore, only CRMP-2B is expressed by oligodendrocytes, cells forming the myelin sheath in the CNS, while both CRMP-2 isoforms are present in axons (Bretin et al. 2005). Proteolytic cleavage of the N-terminus in CRMP-2A also occurs after the activation of NMDA receptors (Bretin et al. 2006).
The C-terminal ‘tail’ region of the CRMPs has no detectable sequence homology to other proteins, it has a high pI of 10.8 and a high content of serine residues and it is predicted to be unfolded in secondary structure predictions (not shown). These attributes resemble those usually associated with the tau/MAP microtubule-binding proteins. In a recent study, a C-terminally truncated form of CRMP-2 was shown to be specific to the PNS; the expression of this form was decreased after sciatic nerve injury (Katano et al. 2006). It has also been shown that the CRMP C-terminal tail is cleaved off by proteolysis both in vitro and in vivo (Deo et al. 2004; Bretin et al. 2006), and it was, thus, not included in our construct. However, it is worth mentioning that important functional sites reside in this domain.
Collapsin response mediator protein 2 is regulated by phosphorylation by a number of kinases, and all phosphorylation sites, with one exception, Ser465, have been mapped to the C-terminal tail (Arimura et al. 2000; Gu et al. 2000; Brown et al. 2004; Cole et al. 2004; Yoshimura et al. 2005). Phosphorylation inhibits the binding of CRMP-2 to tubulin and prevents its effect on tubulin polymerisation (Arimura et al. 2005). It is likely that the C-terminal tail plays a major role in interactions between CRMP-2 and tubulin. Furthermore, the interaction between CRMP-2 and the tetratricopeptide repeat (TPR) domain of kinesin seems to be mediated by residues 440–572, i.e. the C-terminal tail (Kimura et al. 2005).
Collapsin response mediator protein 2 is also O-glycosylated (Cole and Hart 2001; Kanninen et al. 2004), and prediction of glycosylation sites using the NetOGlyc server (Hansen et al. 1998) strongly suggests that the glycosylation site(s) also are in the C-terminal tail (not shown). The effects of glycosylation on CRMP-2 function have not been defined, but it is possible that glycosylation regulates the interactions of CRMP-2 with its ligands and/or its phosphorylation.
Interestingly, a recent screen for brain calmodulin-binding proteins identified CRMP-2 as a putative CaM target (Zhang et al. 2006). Searching for putative CaM-binding sequences (Yap et al. 2000) of CRMP-2, the most likely region would be the last helix of the current structure, residues 475–489. The helix is involved in the tetramerisation interface, not being completely buried, however. Another putative CaM-binding region could be the C-terminal tail. While further studies on CRMP-2–CaM interactions are clearly required, it can be speculated that CaM binding could affect CRMP-2 tetramerisation, its interactions with other proteins, and/or its post-translational modifications.
Previously mapped functional domains and their relation to the three-dimensional structure of CRMP-2
A number of studies have been carried out to find interaction partners for CRMP-2. These studies have, to a large extent, used truncation mutants in cell culture experiments, which have also allowed the mapping of binding domains (Fig. 7b). Apart from a short region around residue 320, none of the studied segments are involved in tetramerisation interfaces. The interaction data can now be discussed in light of the current structural information.
Collapsin response mediator protein 2 binds tubulin and regulates microtubule assembly (Gu and Ihara 2000; Fukata et al. 2002). The functional domain in this case was mapped to the region 323–381, based on a series of truncation mutants. This region was the most effective in promoting microtubule assembly in vitro; it was not enough for in vivo function, however (Fukata et al. 2002). Judging from our structure (Fig. 7b), this region is mostly buried within the fold of CRMP-2 (indeed forming an α/β unit of the TIM barrel), and it is likely that the effects observed with either only this domain present or with CRMP-2 lacking this region, are affected by folding defects.
Collapsin response mediator protein 2 also interacts with Numb, a protein involved in endocytosis (Nishimura et al. 2003). It was suggested that the region 275–322 is required for the binding, based on the behaviour of various truncation mutants. On the basis of the crystal structure, the inactive 1–275 mutant used in the interaction study is most likely misfolded, as it contains a truncated TIM domain, and the recorded lack of activity should therefore be interpreted with caution. The active 1–381 mutant, however, contains the entire TIM barrel and most of the small domain. A direct interaction has also been characterised between CRMP-2 and phospholipase D2, requiring residues 243–300 of CRMP-2 (Lee et al. 2002). This region overlaps with that suggested to bind Numb, and forms a central part of the folded TIM barrel domain.
In summary, while the actual interactions of full-length CRMP-2 are not brought into question by the crystal structure of human CRMP-2 presented in this article, the detailed mapping data from truncation experiments are very difficult to rationalise while analysing the three-dimensional structure. This is, of course, a general problem while mapping binding sites by using linear truncation mutants, when the three-dimensional structure of the studied protein is not accurately known. Based on our work, structure-based experiments can now be carried out on CRMP-2 and its interaction properties.