Structural Requirements for Association of Neurofascin with Ankyrin*

This paper presents the first structural analysis of the cytoplasmic domain of neurofascin, which is highly conserved among the L1CAM family of cell adhesion molecules, and describes sequence requirements for neurofascin-ankyrin interactions in living cells. The cytoplasmic domain of neurofascin dimerizes in solution, has an asymmetric shape, and exhibits a reversible temperature-dependent β-structure. Residues Ser56–Tyr81 are necessary for ankyrin binding but do not contribute to either dimerization or formation of structure. Transfected neurofascin recruits GFP-tagged 270-kDa ankyrinG to the plasma membrane of human embryo kidney 293 cells. Deletion mutants demonstrate that the sequence Ser56–Tyr81 contains the major ankyrin-recruiting activity of neurofascin. Mutations of the FIGQY tyrosine (Y81H/A/E) greatly impair neurofascin-ankyrin interactions. Mutation of human L1 at the equivalent tyrosine (Y1229H) is responsible for certain cases of mental retardation (Van Camp, G., Fransen, E., Vits, L., Raes, G., and Willems, P. J. (1996) Hum. Mutat. 8, 391). Mutations F77A and E73Q greatly impair ankyrin binding activity, whereas mutation D74N and a triple mutation of D57N/D58N/D62N result in less loss of ankyrin binding activity. These results provide evidence for a highly specific interaction between ankyrin and neurofascin and suggest that ankyrin association with L1 is required for L1 function in humans.

This paper presents the first structural analysis of the cytoplasmic domain of neurofascin, which is highly conserved among the L1CAM family of cell adhesion molecules, and describes sequence requirements for neurofascin-ankyrin interactions in living cells. The cytoplasmic domain of neurofascin dimerizes in solution, has an asymmetric shape, and exhibits a reversible temperature-dependent ␤-structure. Residues Ser 56 -Tyr 81 are necessary for ankyrin binding but do not contribute to either dimerization or formation of structure. Transfected neurofascin recruits GFP-tagged 270-kDa ankyrin G to the plasma membrane of human embryo kidney 293 cells. Deletion mutants demonstrate that the sequence Ser 56 -Tyr 81 contains the major ankyrin-recruiting activity of neurofascin. Mutations of the FIGQY tyrosine (Y81H/A/E) greatly impair neurofascin-ankyrin interactions. Mutation of human L1 at the equivalent tyrosine (Y1229H) is responsible for certain cases of mental retardation (Van Camp, G., Fransen, E., Vits, L., Raes, G., and Willems, P. J. (1996) Hum. Mutat. 8, 391). Mutations F77A and E73Q greatly impair ankyrin binding activity, whereas mutation D74N and a triple mutation of D57N/D58N/D62N result in less loss of ankyrin binding activity. These results provide evidence for a highly specific interaction between ankyrin and neurofascin and suggest that ankyrin association with L1 is required for L1 function in humans.
L1, CHL1, neurofascin, NrCAM, and NgCAM in vertebrates and neuroglian in Drosophila are members of the L1CAM family of cell adhesion molecules (1,2). These proteins possess variable ectodomains that engage in homophilic as well as heterophilic interactions and have in common a conserved cytoplasmic domain that binds to the membrane skeletal protein ankyrin (1,3,4). L1CAM family members are abundant in brain tissue (3,4) and participate in diverse cellular activities including axon fasciculation, myelination, synaptogenesis, and axonal guidance (1,5,6). Mutations in the human L1 gene are responsible for developmental abnormalities including mental retardation and hydrocephalus (7)(8)(9).
The cytoplasmic domains of L1CAM cell adhesion molecules contain a highly conserved sequence that has been identified as a binding site for members of the ankyrin family of membrane skeletal proteins (3, 4, 10 -12). Association of L1CAM molecules with ankyrin was first characterized for neurofascin, which was originally identified as a brain protein that associ-ated with ankyrin-coupled affinity columns (3). Subsequent studies demonstrated that other L1CAM family molecules including L1, NrCAM, and neuroglian also have ankyrin binding activity in their cytoplasmic domains (4,10,12). The membrane binding domain of ankyrin has been demonstrated to have two distinct binding sites for neurofascin and other membrane proteins and is proposed to form lateral complexes between ion channels and cell adhesion molecules as well as to couple these proteins to the spectrin-based membrane skeleton (13)(14)(15). Physiologically relevant sites for interactions between ankyrin and L1CAM molecules include nodes of Ranvier and axon initial segments, where neurofascin and NrCAM are concentrated and co-localized with specialized isoforms of ankyrin, 270/480-kDa ankyrin G (16). In addition, L1 and ankyrin B are co-localized and believed to associate with each other in unmyelinated axons based on loss of L1 in premyelinated axon tracts of ankyrin B (Ϫ/Ϫ) mice. 1 Ankyrin binding activity of neurofascin requires a highly conserved sequence from Ser 56 to Tyr 81 in the cytoplasmic domain and in particular the sequence FIGQY (Phe 77 -Tyr 81 ) that is present in all L1CAM family members (11,12). Internal deletion of these sequences abolishes ankyrin binding. In addition, the tyrosine residue in the FIGQY sequence has been identified as the major tyrosine phosphorylation site of neurofascin (11). Phosphorylation of the FIGQY tyrosine eliminates ankyrin binding of neurofascin in vitro and reduces coupling of neurofascin to the cytoskeleton in vivo (11). Inhibition of neurofascin-ankyrin interaction by tyrosine phosphorylation also results in dissociation of cell-cell adhesion mediated by homophilic interactions between neurofascin molecules expressed in cultured cells (18).
Questions remaining unanswered concern the oligomeric state and folding of the neurofascin cytoplasmic domain and the influence of tyrosine phosphorylation and mutations on these parameters. Furthermore, since previous studies on neurofascinankyrin interaction are largely based on in vitro binding assays (3,4,11,14), it is not clear whether the high or low affinity interactions identified in vitro actually occur in living cells.
This study presents analysis of primary and secondary structural requirements for association between neurofascin and 270-kDa ankyrin G. Ankyrin-neurofascin interactions were detected based on ability of transfected neurofascin to recruit GFP-tagged 270-kDa ankyrin G to the plasma membrane of human embryo kidney 293 cells. Results of this assay demonstrate that the FIGQY tyrosine residue as well as certain other aromatic and negatively charged residues in the sequence Ser 56 -Tyr 81 are essential for the neurofascin-ankyrin interaction. Moreover, mutation of the tyrosine residue in the FIGQY sequence to histidine (Y81H) greatly impairs ankyrin binding * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ To whom correspondence should be addressed. activity of neurofascin. An equivalent mutation in the human L1 gene (Y1229H) results in mental retardation and hydrocephalus (19). These results provide evidence for a highly specific interaction between ankyrin and neurofascin involving residues conserved among L1CAM family members and suggest that ankyrin association with L1 is required for L1 function in humans.

MATERIALS AND METHODS
Preparation of cDNA Constructs-Assembly of the full-length cDNA construct of 270-kDa ankyrin G (Ank270) was achieved by ligating the first half of the membrane-binding domain, which was isolated from an adult rat brain 5Ј-stretch plus cDNA library (CLONTECH) by PCR 2 and con- 2 The abbreviations used are: PCR, polymerase chain reaction; HA, hemagglutinin; NF, neurofascin; GFP, Green fluorescent protein.
FIG. 1. The cytoplasmic domain of neurofascin is predominantly a random coil with some temperature-dependent ␤-structure that is unrelated to the sequence from Ser 56 to Tyr 81 . Circular dichroism spectra of the bacteria-expressed cytoplasmic domain of native neurofascin (A) or mutant neurofascin with deletion of the sequence from Ser 56 to Tyr 81 were measured at 5, 37, and 65°C. The dominant negative peak at about 200 nm is indicative of a random coil. The 5°C wavelength scans were subtracted from the 65°C scans to determine the heat-dependent ␤-structure, which is reflected by a positive peak at 195 nm and a negative peak at 220 nm (C). Circular dichroism experiments were performed on an Aviv 62 DS instrument. The samples were dialyzed into buffer containing 5 mM sodium phosphate and 0.5 mM sodium azide at pH 7.4. Wavelength scans were performed from 180 to 253 nm at 5, 37, and 65°C. These results were repeated in three separate experiments. The identity of expressed neurofascin cytoplasmic domain polypeptides was confirmed by N-terminal sequencing (see "Materials and Methods").   Fig. 2) was prepared by ligating the EcoRI-EcoRV fragment of construct Ank270 into the EcoRI-EcoRV sites of construct Ank-Ct. The resultant construct (Ank270-GFP; Fig. 2) is driven by the cytomegalovirus promoter. The full-length rat neurofascin cDNA with an HA epitope at the N terminus (11) was cut out by HindIII-NotI digestion of PBluescript KS vector (Stratagene) and subcloned into the corresponding sites of pEGFP-N1 vector (CLONTECH). The resultant construct (HA-NF; Fig. 1) does not contain the EGFP sequence and is driven by a cytomegalovirus promoter. All other neurofascin constructs were prepared from construct HA-NF unless indicated. The cytoplasmic domaindeleted neurofascin HA-NF(⌬E21-A109) was made by replacing the ScaI-NotI fragment of construct HA-NF with a PCR-amplified fragment containing the sequence of the ScaI-ApaI fragment with a stop codon following the ApaI site (Fig. 2). The extracellular domain-truncated neurofascin (HA-NF(⌬EC); Fig. 2) was prepared through two steps. First, the 5Ј-untranslated region of neurofascin, the start codon, the signal peptide, and HA tag were PCR-amplified and subcloned into the BglII-HindIII Point mutations of the cytoplasmic domain of neurofascin were prepared using the QuickChange site-directed mutagenesis kit (Stratagene) and confirmed by DNA sequencing (Fig. 5). The ApaI-NotI fragment of HA-NF was subcloned into pBluescript vector, in which site-directed mutagenesis was carried out. The confirmed sequences with single amino acid mutations were subcloned back into the exact sites of construct HA-NF from which they had been cut out.
cDNA Transfection and Immunofluorescence-Human kidney 293 cells were cultured in 10% fetal bovine serum in Dulbecco's modified Eagle's medium and transfected with LipofectAMINE in Opti-MEM serum-free medium following the manufacturer's protocol (Life Technologies, Inc.). The amount of each species of co-transfected cDNAs was adjusted to a 1:1 molar ratio. In most cases, 1 g of neurofascin cDNA or its mutants and 1.5 g of 270-kDa ankyrin G cDNA were applied for each co-transfection experiment in a 35-mm culture dish. All co-transfection experiments and subsequent immunofluorescence experiments were performed under the same conditions.
For immunofluorescence, transfected 293 cells were fixed in 2% paraformaldehyde for 10 min and then incubated with blocking buffer (10% normal goat serum and 2% bovine serum albumin in phosphatebuffered saline) for another 5 min before applying the primary antibody against HA epitope (Babco). After 3 h of incubation with the primary antibody, cells were washed three times with phosphate-buffered saline and subjected to secondary stain using tetramethylrhodamine isothiocyanate-conjugated goat anti-mouse antibody for 1 h. All immunostaining procedures were performed at room temperature. In the case of co-transfection of Ank270-GFP and Epo-NF(E21-A109), 0.1% Triton X-100 was included in the blocking buffer, and the primary stain was an antibody specific for the FIGQY sequence in the cytoplasmic domain of neurofascin. All immunofluorescence experiments were recorded using a Zeiss LSM 410 confocal microscope. Experiments were replicated at least three times.
Determination of Physical Properties of the Cytoplasmic Domain of Neurofascin-DNA constructs of the neurofascin cytoplasmic domain were amplified by PCR using 186-kDa rat neurofascin cDNA as a template (3) with added 5Ј NcoI and a 3Ј XhoI restriction sites for subcloning. The PCR products were restricted and ligated into a Pet vector with a C-terminal histidine tag (Novagen Pet 28b(ϩ)). Cytoplasmic domain lacking the histidine tag was also expressed and exhibited the same properties as His tag constructs. Plasmids were transformed into BL21 DE3 pLysS bacteria and expressed with isopropyl-1-thio-␤-D-galactopyranoside induction. Expressed proteins with histidine tags were purified from the soluble fraction of lysed bacteria using a nickel-FIG. 3. Endogenous spectrin of 293 cells is insufficient to recruit transfected ankyrin to the plasma membrane. A, immunostaining of endogenous spectrin in 293 cells, which is predominantly localized at the plasma membrane. B, immunostaining of endogenous ankyrin, which is also concentrated at the plasma membrane. C, transfected 270-kDa ankyrin G (Ank270-GFP) visualized by GFP signal is distributed throughout the cytoplasm of 293 cells.
nitrilotriacetic acid-agarose affinity column (Qiagen). The eluate from the nickel column was then dialyzed and applied to a Mono S high pressure liquid chromatography ion exchanger column, which was subsequently eluted with a sodium chloride salt gradient. Proteins lacking a histidine tag were isolated using Mono Q ion exchange chromatography and gel filtration on Superose 12 in addition to Mono S ion exchange chromatography. The identity of purified polypeptides was established by N-terminal sequencing following transfer to polyvinylidene difluoride paper (3,13). The purity of polypeptides was at least 90% based on SDS-polyacrylamide gel electrophoresis.
The Stokes radii (R s ) of cytoplasmic domains were estimated by gel filtration on a Superose 12 column equilibrated with 10 mM sodium phosphate, 100 mM NaCl, 0.5 mM dithiothreitol, 1 mM NaN 3, pH 7.4, and calibrated with the following protein standards: ferritin (R s ϭ 6.1 nm), catalase (R s ϭ 5.2 nm), bovine serum albumin (R s ϭ 3.5 nm), ovalbumin (R s ϭ 3.05 nm), and cytochrome c (R s ϭ 2 nm). The sedimentation coefficients were determined by sedimentation equilibrium on a Beckman XL-A Optima analytical centrifuge, and the sedimentation patterns were analyzed by the Ideal 1 software program from Beckman. The measurements of Table I were repeated three times with comparable results. Values are presented for a single representative experiment.

Secondary Structure and Oligomeric State of the Cytoplasmic
Domain of Neurofascin-The secondary structure of the cytoplasmic domain of native neurofascin expressed in bacteria was evaluated by CD spectroscopy (Fig. 1). At 5°C, the dominant negative peak at 200 nm in the CD spectrum (Fig. 1A) suggested the native cytoplasmic domain is predominantly a random coil (21). Hydrodynamic measurements of the Stokes radius (R s ϭ 3.1 nm) and the frictional ratio (f/f 0 ϭ 1.7) also indicate an extended conformation of the cytoplasmic domain of native neurofascin (Table I). At higher temperatures (37 and 65°C), however, there is evidence of structure, as demonstrated by the reduction of absorbance above 210 nm at higher temperatures (Fig. 1A). Difference spectra between 65 and 5°C revealed a curve consistent with ␤-structure with a positive peak at 195 nm as well as a negative peak at 220 nm (Fig. 1C). The appearance of ␤-structure is fully reversible upon lowering the temperature (data not shown). This temperature dependence implies hydrophobic interactions involved in stabilizing ␤-structure in the cytoplasmic domain of neurofascin (21).
The sequence Ser 56 -Tyr 81 is necessary for ankyrin binding to intact neurofascin expressed in neuroblastoma cells (11). The contribution of residues Ser 56 -Tyr 81 to the secondary structure of the cytoplasmic domain of neurofascin expressed in bacteria was evaluated by measuring the CD spectrum of the cytoplasmic domain with internal deletion of Ser 56 -Tyr 81 (Fig. 1B). The appearance of a positive peak at 195 nm and a negative peak at 220 nm in the difference spectrum, which is almost identical to that of native neurofascin, indicates that the sequence Ser 56 -Tyr 81 does not contribute to formation of the temperature-dependent secondary structure.
The cytoplasmic domain of neurofascin exists as a dimer or in a monomer-dimer equilibrium, based on molecular weight determined by sedimentation equilibrium measurements (Table I). The cytoplasmic domain with internal deletion of Ser 56 -Tyr 81 also behaves as a dimer (Table I). Sedimentation measurements were performed at 25°C, which is a condition with minimal temperature-induced secondary structure detected by CD spectroscopy. Therefore, assembly of neurofascin cytoplas- mic domains into dimers does not require the thermally induced folding detected in Fig. 1. The cytoplasmic domain contains only minimal predicted coiled-coil ␣-helix, which is consistent with CD data but leaves unexplained the basis for dimerization.
These data demonstrate that, although the cytoplasmic domain of neurofascin is predominantly a random coil, it is capable of forming a dimer and ␤-structure at physiological temperatures. However, there is no evidence that the sequence from Ser 56 to Tyr 81 , which is essential for ankyrin binding to fulllength neurofascin (11), contributes to dimerization or to formation of ␤-structure. The sequence Ser 56 -Tyr 81 thus has little secondary structure, although this segment could be configured as a loop or turn that is not easily detected by CD spectroscopy.
Development of an Assay to Evaluate Neurofascin-Ankyrin Interactions in Living Cells-To evaluate interactions between ankyrin and neurofascin in cells, we developed an assay system based on co-transfection of GFP-tagged 270-kDa ankyrin G and HA-tagged neurofascin (Fig. 2) into human embryonic kidney cells (293 cells). Endogenous spectrin and endogenous ankyrin were concentrated at the plasma membrane of 293 cells (Fig.  3). However, transfected GFP-tagged 270-kDa ankyrin G (Ank270-GFP) was distributed throughout the cytoplasm (Fig.  3). GFP-tagged ankyrin was recruited to the plasma membrane when co-transfected with neurofascin (HA-NF) (Fig. 4A). The ability of neurofascin to recruit ankyrin to the plasma membrane did not require the extracellular domain of neurofascin. Deletion of the extracellular domain (HA-NF(⌬EC)) had little effect on recruitment of co-transfected ankyrin to the plasma membrane (Fig. 4C).
The cytoplasmic domain of neurofascin was necessary and sufficient to provide effective membrane-binding sites for recruitment of transfected ankyrin to the plasma membrane (Fig.  4, B and D). Deletion of a major portion of the cytoplasmic domain of neurofascin extending from Glu 21 to the C terminus residue Ala 109 (HA-NF(⌬E21-A109); Fig. 5) abolished ankyrin recruitment to the plasma membrane (Fig. 4B). The numbers assigned to the amino acid residues in the cytoplasmic domain of neurofascin are based on the convention where the first amino acid following the transmembrane domain (the lysine residue in the sequence KRSRGG; see Fig. 5) is residue 1 (11). The addition of the sequence from Glu 21 to Ala 109 to the cytoplasmic domain-deleted erythropoietin receptor resulted in a chimeric protein (Epo-NF(E21-A109)) with full activity in recruiting co-transfected ankyrin to the plasma membrane (Fig.  4D). These data indicate that the sequence from Glu 21 to Ala 109 of the cytoplasmic domain of neurofascin determines activity in binding to ankyrin.
The Sequence Ser 56 -Tyr 81 Contains the Major Ankyrin-binding Site of Neurofascin-The in vivo assay described above was used to evaluate the effect on ankyrin binding of various internal deletions in the cytoplasmic domain of neurofascin. Internal deletion from Ser 56 to Tyr 81 (HA-NF(⌬S56-Y81)) ( Fig. 5) abolished recruitment of co-transfected ankyrin to the plasma membrane in the in vivo assay (Fig. 6A), as observed previously in direct binding assays (11). Two other mutants with shorter internal deletions, one from Gln 70 to Tyr 81 (HA-NF(⌬Q70-Y81)) and the other from Phe 77 to Tyr 81 (HA-NF(⌬F77-Y81)) (Fig. 5), also eliminated the ability of neurofascin to recruit co-transfected ankyrin to the plasma membrane (Fig. 6, B and C, respectively). Neurofascin with internal deletion of Phe 77 -Tyr 81 retained about 20% ankyrin binding activity compared with the native protein in in vitro binding assays (11). However, the cell assay developed in this study failed to detect this residual ankyrin binding activity (Fig. 6C).
Residues in the cytoplasmic domain of neurofascin preceding Ser 56 and following Tyr 81 do not contribute significantly to ankyrin binding activity. As demonstrated above using constructs Epo-NF(E21-A109)) and HA-NF(⌬E21-A109) (Fig. 5), the sequence before Glu 21 that includes the extracellular and transmembrane domains and part of the cytoplasmic domain (K1-P20) was neither necessary (Fig. 4D) nor sufficient (Fig.  4B) for recruitment of ankyrin to the plasma membrane. A neurofascin mutant with an internal deletion of the sequence from Glu 21 to Glu 55 (HA-NA(⌬E21-E55)) retained full activity to recruit transfected ankyrin to the plasma membrane (Fig.  7A). Deletion of the sequence following Tyr 81 (HA-NA(⌬T82-A109)) also exerted little effect on recruitment of ankyrin to the plasma membrane (Fig. 8A).
While residues outside of the critical Ser 56 -Tyr 81 stretch are not required for recruitment of ankyrin to the plasma membrane, sequences within Ser 56 -Tyr 81 are essential for activity. Internal deletions extending from Glu 21 to Gly 64 (HA-NA(⌬E21-G64)) or from Glu 21 to Gln 70 (HA-NA(⌬E21-Q70)) ( Fig. 5) greatly impair ankyrin-recruiting activity (Fig. 7, B and  C, respectively). Deletion of residues from Phe 77 to Ala 109 (HA-NA(⌬F77-A109)) ( Fig. 5) totally abolished ankyrin recruitment to the plasma membrane (Fig. 8C). Taken together, the above data indicate that the sequence from Ser 56 to Tyr 81 contains the major ankyrin-binding site in the cytoplasmic domain of neurofascin.
The FIGQY Tyrosine (Tyr 81 ) Is Essential for Neurofascin-Ankyrin Interactions-The neurofascin mutant with deletion of the sequence following Tyr 81 (HA-NA(⌬T82-A109); Fig. 5) retains full activity in recruitment of transfected ankyrin to the plasma membrane (Fig. 8A). However, if the deletion extends one additional residue to include the tyrosine residue Tyr 81 (HA-NA(⌬Y81-A109); Fig. 5), recruitment of transfected ankyrin to the plasma membrane is almost abolished (Fig. 8B). These experiments demonstrated a critical role for the FIGQY tyrosine in neurofascin-ankyrin interactions.
Mutations of Tyr 81 were used to further evaluate the structural requirements at the FIGQY tyrosine site. The mutation Y81F (HA-NF(Y81F)) retains full ankyrin-recruiting activity (Fig. 9A), suggesting the hydroxyl group of the tyrosine residue is not directly involved in binding. Previous studies have demonstrated that phosphorylation of Tyr 81 at this hydroxyl group abolishes ankyrin binding activity (11,18). Mutation of Tyr 81 to glutamic acid (HA-NF(Y81E) mimics phosphorylation by intro-ducing negative charges and greatly impairs recruitment of ankyrin to the plasma membrane (Fig. 9C). Mutation of Tyr 81 to alanine (HA-NF(Y81A) totally abolishes ankyrin-recruitment (Fig. 9B), suggesting that the aromatic ring of Tyr 81 also is involved in neurofascin-ankyrin interaction.
The clinical importance of the tyrosine residue in the FIGQY sequence is emphasized by the finding that a mutation in the cytoplasmic domain of human L1 molecule with the corresponding tyrosine residue changed to histidine (Y1229H) is responsible for the development of mental retardation and hydrocephalus (19). An equivalent mutation in neurofascin (Y81H) was prepared and evaluated for its activity to recruit co-transfected ankyrin to the plasma membrane. Mutation of Tyr 81 to histidine greatly impairs ankyrin recruitment activity in 293 cells (Fig. 9D). This result suggests that disruption of the interaction between L1 and ankyrin may be the molecular basis for symptoms of patients with the L1 Y1229H mutation.
Interaction between Neurofascin and Ankyrin Involves Conserved Aromatic and Negatively Charged Residues within the Gln 70 -Tyr 81 Stretch-Other aromatic residues in addition to Tyr 81 in the highly conserved Q70FNEDGSFIGQY81 sequence also participate in association between neurofascin and ankyrin. F77A (HA-NF(F77A)) and F71A (HA-NF(F71A)) ( Fig.  5) mutations both impair ankyrin recruitment to the plasma membrane (Fig. 10, A and B, respectively). Moreover, mutation F71L (HA-NF(F71L)), compared with mutation F71A, does not significantly improve ankyrin recruitment to the plasma membrane (Fig. 10C), suggesting that the function of the aromatic group in neurofascin-ankyrin interactions cannot be replaced by a hydrophobic side chain.
Electrostatic interactions between neurofascin and ankyrin have been inferred from in vitro binding assays based on sensitivity to salt (13,14). Neutralizing a negative charge by mutation of Glu 73 to glutamine (HA-NF(E73Q)) abolished the ankyrin-recruiting activity of the mutant neurofascin (Fig.  11A). However, mutation of Asp 74 to asparagine (HA-NF(D74N)) had much less effect on neurofascin-ankyrin association (Fig. 11B). Changing all three aspartic acid residues at Asp 57 , Asp 58 , and Asp 62 to asparagine (HA-NF(D57,58,62N) had little effect on recruiting ankyrin to the plasma membrane (Fig. 11C). These data (summarized in Fig. 5) indicate that electrostatic interactions between neurofascin and ankyrin are confined to specific charged residues and provide evidence for a high degree of specificity in the interaction between neurofascin and ankyrin.

DISCUSSION
This paper presents the first structural analysis of the conserved cytoplasmic domain of the L1CAM family of cell adhesion molecules and provides a detailed structural and functional analysis of neurofascin-ankyrin interactions in living cells. The cytoplasmic domain of neurofascin dimerizes in solution, has an asymmetric shape, and exhibits a reversible temperature-dependent ␤-structure. The sequence from Ser 56 to Tyr 81 , which is necessary for ankyrin binding, does not contribute to either dimerization or formation of structure in the cytoplasmic domain of neurofascin. A qualitative assay for evaluation of neurofascin-ankyrin interactions in living cells has been developed. Using this assay, we confirmed results of previous in vitro binding assays in living cells (11) and provided additional evidence that the sequence from Ser 56 to Tyr 81 contains the major ankyrin-binding site. The FIGQY tyrosine (Tyr 81 ), which is the site for tyrosine phosphorylation that abolishes ankyrin binding (11), has been demonstrated to be critical for neurofascin-ankyrin interactions. Mutation of the FIGQY tyrosine to histidine greatly impairs neurofascinankyrin interactions. An equivalent mutation in the human L1 molecule (Y1229H) is responsible for certain cases of hydrocephalus and mental retardation (19). Since L1 and neurofascin share a conserved cytoplasmic binding site for ankyrin (4,11), disruption of L1-ankyrin interactions may be the molecular basis for pathology due to this L1 mutation (Y1229H). Other conserved aromatic and negatively charged residues in the sequence Ser 56 -Tyr 81 are also shown to contribute to neurofascin-ankyrin interactions. Mutation E73Q, which has a minimal alteration of the side chain, greatly impairs ankyrin binding activity, whereas a triple mutation of D57N/D58N/D62N resulted in no loss of ankyrin binding activity. These data provide strong evidence for a high degree of specificity in the interaction between neurofascin and ankyrin.
The new assay for evaluation of neurofascin-ankyrin interactions in co-transfected 293 cells is rapid and effective, although qualitative, and has several potential applications. This assay could be used to screen various point mutations in ankyrin as well as in the L1CAM family of cell adhesion molecules for their effects on ankyrin-L1CAM interactions in living cells. The membrane recruitment assay could also be used to identify potential proteins such as protein-tyrosine kinases and protein-tyrosine phosphatases that are involved in regulation of association of neurofascin with ankyrin (11). The clinical importance of ankyrin-L1CAM interactions has been exemplified by the point mutation Y1229H in the human L1 gene. Moreover, the membrane recruitment assay could be applied to screen pharmaceutical drugs that can modulate ankyrin-L1CAM interactions and perhaps cause developmental defects in the nervous system. Nonetheless, the assay has drawbacks due to lack of quantitative information and does not provide a dissociation constant (K d ) describing the affinity of association of mutant neurofascins with ankyrin. In addition, without a careful control of the amount of cDNAs used in co-transfection experiments, one species of cDNA could be overexpressed and complicate subsequent analysis. By using the same type of vector, the same promoter and 3Ј-untranslated region, and carefully controlled cDNA ratios, our results can be successfully reproduced.
Previous studies of ankyrin-neuroglian interactions in Drosophila S2 tissue culture cells have demonstrated that ankyrin selectively associates with neuroglian at sites of cell-cell contact but not other regions of the plasma membrane (10,12). These observations suggest that recruitment of ankyrin by neuroglian requires activation of neuroglian by extracellular interactions. However, our study demonstrates that recruitment of transfected ankyrin to the plasma membrane of cultured human embryonic kidney cells (293 cells) is solely governed by the presence of the cytoplasmic domain of neurofascin localized at the plasma membrane and is independent of whether the co-transfected cell is isolated or has contact with other cells. The basis for these differences may reflect specialized behavior among L1 CAM family members, differences between S2 cells and human 293 cells, and differences between 270-kDa ankyrin G and the ankyrin expressed in S2 cells.
An unexpected result of this study is identification of the cytoplasmic domain of neurofascin as a dimer in solution. The effects of dimerization of neurofascin on ankyrin binding and on lateral organization of neurofascin in the plane of the membrane have yet to be determined. Since the ankyrin-binding sequence from Ser 56 -Tyr 81 does not contribute to dimerization and sequences outside the Ser 56 -Tyr 81 sequence do not contribute to ankyrin binding, it is likely that formation of neurofascin dimers does not directly affect ankyrin binding, at least at the qualitative level detectable in the membrane recruitment assay. However, dimerization of neurofascin provides a potential mechanism for formation of neurofascin-ankyrin polymers. Considered together with the multiple binding sites for neurofascin in ankyrin R (13,14), the existence of neurofascin dimers implies that neurofascin and ankyrin are able to form lateral complexes containing multiple copies of neurofascin, ankyrin, and possibly other ankyrin-binding membrane proteins. These complexes could be further immobilized by coupling to the spectrin-based membrane skeleton through the spectrin-binding domain of ankyrin. The ability to form such large immobilized complexes between ankyrin and neurofascin could be important for the assembly of specialized membrane domains such as axon initial segments and nodes of Ranvier, where these proteins are localized (16). 3 Lack of discernible secondary structure in the sequence Ser 56 -Tyr 81 indicates that effects of various deletions and mu-tations in this sequence on ankyrin binding are more likely exerted by direct perturbation of the contact site with ankyrin rather than by disrupting the conformation of the cytoplasmic domain of neurofascin. Atomic structure of ankyrin repeats of the transcription factor GAPB-␤ revealed that the protein binding site in ankyrin repeats is configured as a tandem array of extended loops, where the tips of these loops provide the binding interface (17). Extrapolation of this information regarding the ankyrin repeats of GAPB-␤ to ankyrin suggests the sequence Ser 56 -Tyr 81 in the cytoplasmic domain of neurofascin also is in contact with predicted loops. This prediction ulti- 3 11. Conserved negative charges are involved in neurofascin-ankyrin association. A, mutation E73Q greatly diminishes the ankyrin recruiting activity. B, mutation D74N has a moderate deteriorating effect on membrane recruitment of co-transfected Ank270-GFP. C, mutation with Asp 57 , Asp 58 , and Asp 62 changed to asparagine keeps full ankyrin-recruiting activity. 1 and 1Ј, 2 and 2Ј, and 3 and 3Ј represent the same transfected cells, respectively. 1, 2, and 3 are immunostaining of the HA epitope of mutant neurofascins. 1Ј, 2Ј, and 3Ј are the GFP fluorescence of co-transfected Ank270-GFP. mately can be evaluated by determining the structure of ankyrin-neurofascin complexes. It may also be possible to mutate ankyrin at predicted loop sites and "repair" loss of binding of certain neurofascin mutations as well as create ankyrins specifically lacking neurofascin binding activity.
In summary, this study presents a structural and functional analysis of the interaction between ankyrin and the cytoplasmic domain of neurofascin in living cells. Since the ankyrinbinding site is conserved among members of L1CAM family of cell adhesion molecules, our results with neurofascin are likely to apply to other L1CAM molecules.