Cellular localization, oligomerization, and membrane association of the hereditary spastic paraplegia 3A (SPG3A) protein atlastin.

Hereditary spastic paraplegias comprise a group of clinically heterogeneous syndromes characterized by lower extremity spasticity and weakness, with distal axonal degeneration in the long ascending and descending tracts of the spinal cord. The early onset hereditary spastic paraplegia SPG3A is caused by mutations in the atlastin/human guanylate-binding protein-3 gene (renamed here atlastin-1), which codes for a 64-kDa member of the dynamin/Mx/guanylate-binding protein superfamily of large GTPases. The atlastin-1 protein is localized predominantly in brain, where it is enriched in pyramidal neurons in the cerebral cortex and hippocampus. In cultured cortical neurons, atlastin-1 co-localized most prominently with markers of the Golgi apparatus, and immunogold electron microscopy revealed a predominant localization of atlastin-1 to the cis-Golgi. Yeast two-hybrid analyses and co-immunoprecipitation studies demonstrated that atlastin-1 can self-associate, and gel-exclusion chromatography and chemical cross-linking studies indicated that atlastin-1 exists as an oligomer in vivo, most likely a tetramer. Membrane fractionation and protease protection assays revealed that atlastin-1 is an integral membrane protein with two predicted transmembrane domains; both the N-terminal GTP-binding and C-terminal domains are exposed to the cytoplasm. Together, these findings indicate that the SPG3A protein atlastin-1 is a multimeric integral membrane GTPase that may be involved in Golgi membrane dynamics or vesicle trafficking.

tracts of the spinal cord, which constitute the longest motor and sensory axons in the central nervous system (6). HSPs have historically been classified as "pure" or "uncomplicated" if spastic paraplegia occurs in isolation and "complicated" if other neurological abnormalities are present (7). More recently, the identification of new genetic loci for HSPs has permitted a molecular classification of HSPs (2)(3)(4)(5). Of the 20 known loci (SPG1-20), 11 are autosomal dominant, six are autosomal recessive, and three are X-linked. Although most HSP patients experience progressive worsening of their symptoms, for some, the disorder does not appear to be progressive (3,4). Thus, although some HSPs are clearly neurodegenerative disorders (e.g. SPG4), others such as SPG1 (due to mutations in the L1 cell adhesion molecule) and SPG3A (due to mutations in the atlastin GTPase) may be neurodevelopmental disorders (3,4).
Eight disease genes for HSPs have now been identified; and based on the proteins involved, several mechanisms for pathogenesis have been advanced. These include aberrant cell signaling or migration, abnormalities of mitochondrial chaperones, abnormalities of myelination, and defects in intracellular trafficking and transport (2)(3)(4)(5). Proteins mutated in HSPs that have been implicated in cellular protein or vesicle trafficking include KIF5A (SPG10), spastin (SPG4), spartin (SPG20; Troyer syndrome), and atlastin (SPG3A) (reviewed in Refs. [2][3][4][5]. KIF5A is a neuronal kinesin heavy chain motor protein involved in the transport of macromolecules and membranous organelles along the axon (8). Spastin, a member of the AAA (for ATPases associated with a variety of cellular activities) protein family, associates with microtubules, and spastin overexpression causes microtubule disassembly (9). The spartin protein is similar to the VPS4, SNX15 (sorting nexin-15), and SKD1 proteins, which are involved in endosome morphology and protein trafficking of endosomal compartments (10). These latter proteins share with both spastin and spartin a region called the MIT (contained within microtubule-interacting and trafficking molecules) or ESP (present in End13/VPS4, SNX15, and PalB) domain (2,11). Based on its similarity to members of the dynamin/Mx/guanylate-binding protein (GBP) superfamily of large GTPases (12), the SPG3A protein atlastin (renamed here atlastin-1) has been implicated in intracellular trafficking, yet little is known regarding its cellular localization or function (13).
Among the HSPs, SPG3A is particularly notable for the very early onset of pure spastic paraplegia. Five missense mutations and one single base insertion with premature termination of the predicted 558-amino acid coding region of atlastin-1 have been reported (13)(14)(15)(16). It has been speculated that these mutations may alter the structure, interactions, or GTPase activity of atlastin-1 (13)(14)(15)(16). Of the members of the dynamin/Mx/ GBP superfamily, atlastin-1 is most similar to GBPs. Like GBPs, atlastin-1 possesses an RD loop instead of the classical (N/T)KXD sequence within the third motif of the guanylatebinding consensus triad (13). However, atlastin-1 lacks a Cterminal isoprenylation motif and the C-terminal ␣12/13 helix motif, two characteristic structural features of many of the GBPs (12). Here, we demonstrate that atlastin-1 is an oligomeric GTPase that, unlike GBPs, is composed of subunits that are integral membrane proteins, with both N and C termini exposed to the cytoplasmic compartment. Atlastin-1 localizes prominently to the Golgi apparatus and is enriched in cerebral cortical pyramidal cells, a subpopulation of which exhibit a "long axonopathy" in patients with SPG3A.

EXPERIMENTAL PROCEDURES
Sequence Analysis-Atlastin-1 homologs were identified with BLAST (17). Chromosomal assignments were made using the NCBI Map-Viewer. 2 Transmembrane helical domains were predicted using the SOSUI system (18). Protein alignments were performed with ClustalW (19). Protein sequence similarities were calculated using BESTFIT analysis (Wisconsin Package Version 10.3, Accelrys, San Diego, CA).
Eukaryotic DNA Expression Constructs and Cell Transfection-The full coding sequence of the atlastin-1 GTPase (GenBank TM /EBI accession number NM_015915) was amplified by PCR using Pfu Turbo (Stratagene, La Jolla, CA) from a Marathon human brain (cerebral cortex) cDNA library (Clontech) and confirmed by DNA sequencing. The full-length atlastin-1 cDNA was cloned into the XmaI site of the eukaryotic expression vector pGW1 with Myc or hemagglutinin (HA) epitope tags at the N terminus as described previously (20). The fulllength atlastin-1 cDNA was also cloned into the XmaI site of pRK5 (Genentech, South San Francisco, CA) for expression of the untagged protein. Torsin A (accession number AF007871) was cloned as a HindIII-BglII fragment into pGW1, preceding an in-frame C-terminal Myc tag. Site-directed mutagenesis was performed using the QuikChange method (Stratagene). African green monkey COS-7 cells (American Type Culture Collection CRL-1651) were maintained, transfected, and harvested as described previously (20).
Tissue Preparation and Subcellular Fractionation-Human tissue homogenates were obtained from Clontech. Brain subcellular fractions were prepared from Sprague-Dawley rats (150 -175 g; Charles River Laboratories, Wilmington, MA). Dissected brains were homogenized in 0.32 M sucrose and 10 mM HEPES (pH 7.4) and centrifuged at 1330 ϫ g for 3 min, generating a pellet (P1) and a supernatant (S1). The S1 supernatant was centrifuged at 21,200 ϫ g for 10 min, producing a pellet (P2) and a supernatant (S2). The S2 supernatant was then centrifuged at 200,000 ϫ g for 1 h, generating a P3 pellet and an S3 supernatant. Protein concentrations were determined by the BCA assay (Pierce) with bovine serum albumin as the standard.
Immunohistochemistry-Three adult male Sprague-Dawley rats (150 -200 g) were perfused transcardially under deep pentobarbital anesthesia with 0.9% NaCl in 0.1 M phosphate buffer (pH 7.4), followed by 4% paraformaldehyde and finally 10% sucrose in 0.1 M phosphate buffer. Brains were post-fixed for 1 h and incubated overnight in phosphate-buffered 30% sucrose. Brains were then placed on a freezing microtome and cut into 50-m-thick sections. These sections were collected in phosphate-buffered saline (PBS; pH 7.4) and stored at Ϫ20°C in a cryoprotective solution consisting of 30% sucrose and 30% ethylene glycol in 0.05 M phosphate buffer (pH 7.2) until used.
The immunohistochemical reaction was carried out using the ABC Elite kit (Vector Labs, Inc., Burlingame, CA) according to the manufacturer's instructions. Sections were preincubated in 1.0% normal goat serum (NGS), 10% bovine serum albumin, and PBS for 30 min at 25°C and then incubated for 48 h at 4°C with anti-atlastin-1 antibodies (No. 5409; 0.7 g/ml) in 0.1% NGS, 1% bovine serum albumin, and PBS. In control experiments, the anti-atlastin-1 antibodies (No. 5409) were preincubated with the immunogenic peptide (0.25 M) prior to use. The sections were then rinsed with PBS and incubated with biotinylated anti-rabbit IgG (Vector Labs, Inc.) in PBS with 1.5% NGS for 1 h at 25°C. After rinsing with PBS, sections were incubated with the ABC complex (Vector Labs, Inc.) for 45 min and rinsed again. Peroxidase staining was revealed using 3,3Ј-diaminobenzidine (Fast DAB kit, Sigma). Sections were mounted on Superfrost Plus slides (Fisher), airdried, dehydrated, and coverslipped with Cytoseal 60 (Stephens Scientific, Riverdale, NJ).
Immunogold Electron Microscopy-Rat cortical neuron cultures were prepared as described above. After 6 days in culture, neurons were fixed with 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) for 30 min and then washed with 0.1 M phosphate buffer. The cells were permeabilized and blocked in 5% NGS, 0.1% saponin, and PBS for 1 h and incubated with anti-atlastin-1 antibodies (No. 5409) or without primary antibodies as a control in blocking buffer for 1 h. After washing with 1% NGS in PBS with 2% nonfat milk in PBS, cells were incubated with 1.4-nm Nanogold gold-conjugated anti-rabbit secondary antibodies (1:250 dilution; Nanoprobes, Yaphank, NY) in 2% nonfat milk in PBS for 1 h. After washing with 2% nonfat milk in PBS, the cells were fixed with 2% glutaraldehyde in PBS for 30 min. Finally, the cells were thoroughly washed with PBS and distilled water, silver-enhanced (HQ silver kit, Nanoprobes), and washed again with water and 0.1 M phosphate buffer. The cells were treated with 0.2% OsO 4 in 0.1 M phosphate buffer for 30 min, mordanted en bloc with 0.25% uranyl acetate in acetate buffer (pH 5.0) overnight, washed and dehydrated with serial concentrations of ethanol, and finally infiltrated and embedded in epoxy resins. Thin sections of ϳ70 nm were counterstained with uranyl acetate and lead citrate and examined under a Jeol 1200 EXII transmission electron microscope. Digital images were collected with an XR-100 CCD camera (Advanced Microscopy Techniques, Danvers, MA).
Membrane Association Assays-COS-7 cells overexpressing atlastin-1 were washed twice with 10 mM Tris-HCl (pH 7.5) and then harvested in 10 mM Tris-HCl (pH 7.5), 10 mM NaCl, and 1.5 mM MgCl 2 . After multiple passes through a 25-gauge needle, the homogenate was centrifuged at 1330 ϫ g. The post-nuclear supernatant was sonicated and then recentrifuged at 200,000 ϫ g for 60 min, yielding a pellet (lysed membrane fraction) and a soluble fraction. The membrane fraction was treated with 1 M NaCl and 25 mM phosphate buffer (pH 7.4), with 100 mM glycine buffer (pH 2.8), with 100 mM carbonate buffer (pH 11.0), with 0.1% Triton X-100 and PBS, or with 1.0% sodium deoxycholate and PBS as indicated and centrifuged at 200,000 ϫ g for 60 min 2 Available at www.ncbi.nlm.nih.gov.
to generate a final pellet and supernatant. In other experiments, soluble and membrane fractions were prepared from COS-7 cells overexpressing Myc-tagged wild-type atlastin-1 or various Myc-tagged atlastin-1 deletion constructs. Equal proportions of the soluble and membrane fractions were then immunoblotted with anti-Myc antibodies. Membranes from COS-7 cells overexpressing untagged atlastin-1 and P3 membrane fractions prepared from rat brain were subjected to phase partitioning with Triton X-114 as described by Bordier (24), and equal proportions of the aqueous and detergent phases were immunoblotted with anti-atlastin-1 antibodies (No. 5409).
Yeast Two-hybrid Tests-Yeast two-hybrid tests were performed using the L40 yeast strain harboring the reporter genes HIS3 and ␤-galactosidase under the control of upstream LexA-binding sites as described previously (20). Atlastin-1 deletion constructs were produced by PCR amplification using Pfu Turbo and cloned in-frame into pGAD10 prey and pBHA bait vectors (Clontech). All constructs were confirmed by DNA sequencing. Strength of interaction was assayed by ␤-galactosidase and HIS3 induction as described previously (20).
Immunoprecipitation and Chemical Cross-linking-COS-7 cells cotransfected with HA-and Myc-atlastin-1 or transfected with Myc-atlastin-1 alone were washed twice with PBS and then harvested in 0.5% Triton X-100 and PBS and clarified by centrifugation at 130,000 ϫ g for 30 min. Extracts (100 g of protein) were incubated for 1-2 h at 4°C with 5 g of rabbit polyclonal anti-HA probe antibodies (Y-11) precoupled to protein A-Sepharose CL-4B (Amersham Biosciences). Beads were washed three times with 0.5% Triton X-100 and PBS. Bound proteins were resolved by SDS-PAGE and immunoblotted with mouse monoclonal anti-Myc antibodies. Chemical cross-linking with dithiobis(succinimidyl propionate) (Pierce) was performed using high speed pellets from the post-nuclear supernatant derived from Myc-atlastin-1overexpressing COS-7 cells. Pellets were resuspended in PBS, and dithiobis(succinimidyl propionate) was added for 30 min on ice. Crosslinked products were resolved by SDS-PAGE under nonreducing conditions and immunoblotted with anti-Myc antibodies.
Gel-exclusion FPLC-Myc-tagged wild-type or deletion mutants of atlastin-1 overexpressed in COS-7 cells were lysed in 0.1% Triton X-100 and PBS and clarified by centrifugation at 130,000 ϫ g for 30 min. The soluble extract was applied to a Superdex 200 HR10/30 FPLC column (Amersham Biosciences) at a flow rate of 0.25 ml/min in 0.1% Triton X-100 and PBS. Fractions (0.25 ml) were collected, and proteins were resolved by SDS-PAGE and then immunoblotted using anti-Myc antibodies. Protein standards (Sigma and Amersham Biosciences) in 0.1% Triton X-100 and PBS were applied to the column to generate a standard curve, from which the native molecular masses for wild-type and deletion mutants of atlastin-1 were calculated.

RESULTS
Atlastin Family of GTP-binding Proteins-We searched nucleotide and protein data bases for human proteins similar to atlastin (renamed atlastin-1) and found two highly related proteins (Fig. 1). One, which was previously identified as ARL-6-interacting protein-2 (GenBank TM /EBI accession number NM_022374), we have renamed atlastin-2. The other we have named atlastin-3 (accession number AK097588). Whereas the atlastin-1 gene is found on chromosome 14q22.1 (locus ID 51062), the atlastin-2 gene localizes to chromosome 2p22.3 (locus ID 64225), and the atlastin-3 gene to chromosome 11q13.1 (locus ID 25923). These proteins show extensive homology, with a large central core of higher similarity that includes the GTP-binding region in the N-terminal portion; the most divergence occurs at the C and N termini (Fig. 1). There is 69% amino acid identity and 79% similarity between atlastin-1 and atlastin-2 over the central 533 residues and 66% identity and 75% similarity between atlastin-1 and atlastin-3 over 476 residues. Atlastin-2 and atlastin-3 share 67% identity and 78% similarity over 482 residues. Also, two predicted transmembrane helices (SOSUI) (18) are conserved in all three proteins. Using the same SOSUI prediction program, human GBP1 (accession number NM_002053) and the related GTPase Mag-2 (accession number M81128) lack any predicted transmembrane domains, consistent with previous reports (12). Consensus sequences for N-linked glycosylation and Myb DNA binding previously reported for atlastin-1 (13) are not conserved in atlastin-2 and atlastin-3 (Fig. 1).
Identification and Subcellular Localization of the Atlastin-1 Protein-We generated specific anti-peptide antibodies against the divergent N (No. 5409) and C (No. 4705) termini of atlastin-1. Atlastin-1 was detected at ϳ64 kDa on immunoblots of homogenates from atlastin-1-overexpressing COS-7 cells, consistent with its predicted size, but not in those from untransfected cells ( Fig. 2A). Similar results were obtained with antipeptide antibodies raised against both the N ( Fig. 2A) and C (data not shown) termini. Atlastin-1 was also detected in tissue homogenates from rat and human brain, with no cross-reacting protein bands (Fig. 2A). The immunoreactive signal was abolished when the antibodies were preadsorbed with the immunogenic peptide (1 M) (data not shown). Atlastin-1 was most enriched in brain, but was also present in several other human tissues at much lower levels (Fig. 2B). By comparison, the mitochondrial intermembrane space protein DDP1/TIMM8a had a more homogeneous tissue distribution (Fig. 2B). We examined the subcellular localization of atlastin-1 using rat brain fractions prepared by differential centrifugation (Fig.  2C). Both atlastin-1 and the endoplasmic reticulum protein calregulin were enriched in the microsomal P3 fraction. In contrast, the mitochondrial protein OPA1/Mgm1 was most abundant in the P2 pellet, with much lower levels in the P3 fraction.
We examined the localization of endogenous atlastin-1 in rat brain sections using antibodies generated against the N terminus of atlastin-1 (No. 5409) (Fig. 3). Highly immunopositive cells were most abundant in lamina V of the cerebral cortex, including the motor cortex. Although the neuronal soma was most strongly labeled, the dendritic tree was also stained (Fig.  3, A-C). Immunostaining was eliminated when the anti-atlastin-1 antibodies were first preadsorbed with the immunogenic peptide (Fig. 3D). Labeling was also prominent within the hippocampus, mainly in pyramidal neurons in CA1 and CA3 (Fig. 3, E and F). On the other hand, there was no detectable staining in the cerebellum. Although little immunoreactivity was present in the striatum, with moderate immunostaining limited to a subpopulation of large cholinergic neurons, more robust staining was evident in neurons within the amygdala and several thalamic nuclei. 3 A number of cells in several mesopontine nuclei, including the dorsal substantia nigra pars compacta and the retrorubal area, stained moderately. 3 To establish the localization of atlastin-1 at the subcellular level, cultured cerebral cortical neurons were examined using double-label immunofluorescence. Atlastin-1 staining was found most prominently in the cell soma, with weaker staining in neuronal processes, comprising both axons and dendrites (Fig. 4). Co-localization was very limited with markers for the endoplasmic reticulum (anti-KDEL antibodies), and no overlap of atlastin-1 labeling was seen with mitochondria stained with MitoTracker CMXRos (data not shown). However, p115, a marker for the Golgi apparatus, tightly co-localized with atlastin-1 (Fig. 4, A-C). Co-localization with atlastin-1 was also seen with the Golgi marker GM130, but to a lesser extent (data not shown). The atlastin-1 immunoreactive signal was blocked by preadsorption of the antibodies with the immunogenic peptide (data not shown). At longer exposures, punctate staining was seen in dendrites and axons (Fig. 4, D-F). However, this staining was far less intense than that seen associated with the Golgi apparatus in the neuronal cell body.
To confirm the Golgi predominance of the endogenous atlastin-1 protein, we examined the localization of atlastin-1 in cultured cortical neurons by immunogold electron microscopy 3 B. Lavoie and C. Blackstone, unpublished data. Sm., smooth; Sk., skeletal. C, biochemical fractionation of atlastin-1. Subcellular fractions were prepared from rat brain by differential centrifugation (see "Experimental Procedures" for details). Fractions (20 g of protein/lane) were immunoblotted with antibodies to atlastin-1 (No. 5409), the endoplasmic reticulum protein calregulin (Calreg), and the mitochondrial protein OPA1.

FIG. 1. Atlastin family of proteins in humans.
Atlastin (renamed here atlastin-1 (ATL1)) is highly similar to two proteins we have named atlastin-2 (ATL2; also known as ARL-6-interacting protein 2) and atlastin-3 (ATL3). Residues identical in all three proteins are shaded orange, and those identical in two of the three proteins are shaded green. Bold lines indicate predicted transmembrane domains. GTP-binding motifs are boxed. Asterisks denote sites of missense mutations in patients with SPG3A; these residues are highly conserved in atlastin-2 and atlastin-3. An arrow marks the site of a frameshift mutation in one family with SPG3A. Consensus sites for N-linked glycosylation (#) and the Myb DNA-binding domain (^) in atlastin-1 are indicated; none of these are conserved in both atlastin-2 and atlastin-3. A variant of atlastin-3 with a 26-amino acid insertion following Ser 74 (GenBank TM /EBI accession number CAB56010) and a variant of atlastin-2 with a novel C terminus (38 residues) following Val 545 (accession number AAM97342), both generated by alternative splicing, have also been reported. (Fig. 5). An abundance of immunogold labeling decorated the Golgi apparatus, predominantly the cis-Golgi cisternae (Fig. 5,  A and B). This labeling was not present in control sections in which the primary antibody was omitted (Fig. 5, C and D). No specific staining was seen in mitochondria or in the nucleus.
Atlastin-1 Is an Integral Membrane Protein-Based on the results of subcellular fractionation and immunolabeling, we examined whether atlastin-1 is an integral membrane or tightly membrane-associated Golgi protein, comparing it with known proteins under different membrane extraction conditions. Atlastin-1 overexpressed in COS-7 cells partitioned exclusively to the membrane fraction, which had been lysed by sonication to release soluble luminal proteins and subjected to ultracentrifugation (Fig. 6A). Notably, a significant percentage of the soluble luminal protein calregulin was found in the supernatant, confirming the efficient lysis of microsomes. The integral membrane protein calnexin was exclusively found in the membrane fraction (Fig. 6A). This membrane fraction was subjected to various extraction conditions (i.e. high salt, low pH, high pH, and detergents) to determine whether atlastin-1 is an integral membrane protein. Atlastin-1 partitioned to the supernatant in a manner similar to the integral membrane protein calnexin; neither atlastin-1 nor calnexin was released from the membrane fraction by low or high pH buffers, but both atlastin-1 and calnexin were efficiently solubilized by detergents (Fig. 6A). In contrast, the membrane-associated fraction of calregulin was released from membranes not only by detergents, but also by both high and low pH buffers (Fig. 6A). Phase partitioning with Triton X-114 demonstrated a distribution of atlastin-1 predominantly to the detergent phase, whereas the soluble membrane-associated protein calregulin was found in the aqueous phase (Fig. 6B).
To determine whether the two hydrophobic domains in atlastin-1 represent transmembrane domains, we used several Myc-tagged N-and C-terminal atlastin-1 deletions to assess localization to supernatant or membrane fractions (Fig. 6C). Wild-type atlastin-1 was present in the membrane fraction, and small deletions at the N and C termini did not change the membrane association significantly (Fig. 6C). However, a Cterminal deletion that removed the two predicted transmembrane domains, atlastin-1- (1-447), caused a redistribution of the protein to the supernatant fraction (Fig. 6C), indicating that these domains are required for membrane attachment. A 111-residue C-terminal atlastin-1 fragment containing the predicted transmembrane domains (residues 448 -558) was sufficient for membrane association (Fig. 6C).
To evaluate the transmembrane topology of the atlastin-1 protein, we used intact microsomal vesicles from COS-7 cells overexpressing atlastin-1 in a protease protection assay (Fig.  7A). Immunoreactivity to both N and C termini was lost upon treatment of microsomes with proteinase K, although higher concentrations were required for full removal of the N-terminal domain (Fig. 7A). These results indicate that both N and C termini are exposed to the cytoplasmic face of the membrane. In control experiments, the luminal protein calregulin was completely protected from proteolysis in the same samples (Fig.  7A), even at higher concentrations of protease (data not shown). Consistent with this finding is that mutation of the three consensus sites for N-linked glycosylation in atlastin-1 (N177Q, N236Q, and N436Q) did not alter the migration of atlastin-1 upon SDS-PAGE, nor did treatment with peptide N-glycosidase F (Fig. 7B). In control experiments, the known  glycoprotein torsin A (26,27) overexpressed in COS-7 cells was efficiently deglycosylated by peptide N-glycosidase F (Fig. 7B). Thus, we have no evidence that these consensus sites in atlastin-1 are N-glycosylated in vivo, consistent with our proposed transmembrane topology model (Fig. 7C).
Atlastin-1 Is an Oligomeric Protein-Members of the dynamin/Mx/GBP family form homo-oligomeric protein complexes in vivo, and we sought to determine whether atlastin-1 forms oligomers as well. Using yeast two-hybrid tests, we detected a robust interaction of full-length atlastin-1 with itself (Fig. 8A). We were unable to narrow down the domains required for self-interaction by deletion analyses; interactions were decreased or eliminated with both N-and C-terminal deletions (Fig. 8A). Also, we did not find an interdomain association among several different N-and C-terminal domain constructs (Fig. 8A), although interdomain and intramolecular interactions have been demonstrated for other members of the dynamin/Mx/GBP superfamily (12, 28 -31). In co-immunoprecipitation experiments, Myc-atlastin-1 interacted with HA-atlastin-1 (Fig. 8B), demonstrating self-association of atlastin-1 overexpressed in COS-7 cells. Based on the results of chemical cross-linking experiments, atlastin-1 appears most likely to be a homotetramer, with products at ϳ110, ϳ160, and ϳ230 kDa in addition to the monomer band at 64 kDa (Fig. 8C). Using gel-exclusion FPLC, we found that detergent-solubilized atlas-FIG. 6. Atlastin-1 is an integral membrane protein. A, membrane association of atlastin-1. COS-7 cells were transfected with untagged atlastin-1. Atlastin-1 fractionated in the lysed membrane pellet (Memb), but not the soluble cytosolic fraction (Sol) derived from high speed centrifugation of the post-nuclear supernatant (Homog), as detected by immunoblotting with anti-atlastin-1 antibodies. This pellet was subsequently resuspended in the indicated salts or detergents and then refractionated by high speed centrifugation into the supernatant (S) and membrane pellet (P). Atlastin-1 was released from membranes in a manner similar to the integral membrane protein calnexin, but not to the soluble and membrane-associated protein calregulin. DOC, 1.0% sodium deoxycholate; TX-100, 0.1% Triton X-100. B, Triton X-114 phase partitioning. Membranes from atlastin-1-transfected COS-7 cells and rat brain P3 fractions were partitioned with Triton X-114 (TX-114), and aliquots of the starting material (Input) as well as aqueous and detergent (Triton X-114) phases were immunoblotted with anti-atlastin-1 and anti-calregulin antibodies. Atlastin-1 was enriched in the detergent phase, whereas calregulin was found in the aqueous phase. C, hydrophobic domains of atlastin-1 are required for membrane association. Post-nuclear supernatants from COS-7 cells transfected with Myctagged wild-type atlastin-1 or deletion mutants were fractionated as described for A and immunoblotted with anti-Myc antibodies. Deletion of both predicted transmembrane domains (H), but not smaller N-and C-terminal deletions, markedly reduced membrane association of atlastin-1 (arrow). Boundary amino acids are indicated. tin-1 eluted at a native size of ϳ280 kDa, consistent with the proposed tetrameric structure (Fig. 8D). However, because the cross-linked atlastin-1 product at ϳ230 kDa was weakly visible, and it was difficult to estimate the contribution of detergent to the size estimate for the atlastin-1 oligomer by gelexclusion chromatography, we cannot rule out a trimeric quaternary structure for the atlastin-1 protein in vivo. Interestingly, atlastin-1-(1-516), which contains the putative transmembrane domains, also migrated at a similar native size upon gel-exclusion chromatography. However, atlastin-1-(1-447), which lacks the transmembrane domains, eluted as multiple peaks (Fig. 8D), demonstrating that the two predicted transmembrane domains are required for proper conformation and/or oligomerization of atlastin-1.
Atlastin-1 Is a GTPase-Using purified CBP-atlastin-1 fusion proteins, we examined whether atlastin-1 possesses GTPase activity. Increasing amounts of atlastin-1 protein successively increased the percentage of GTP hydrolysis (Fig. 9). Thus, atlastin-1 not only binds GTP, as shown in previous studies (32), but also functions as a GTPase (Fig. 9). DISCUSSION Mutations in the SPG3A gene cause a very early onset pure HSP (2)(3)(4)(5). In this study, we have found that the SPG3A gene product atlastin-1 is an oligomeric GTPase composed of subunits that are integral membrane proteins, with both C and N termini exposed to the cytoplasm. We have also identified two other members of the atlastin family of GTPases that share significant sequence similarity with atlastin-1 and likely the same structural features: atlastin-2 and atlastin-3 (Fig. 1). These proteins are of similar size and have the same highly conserved GTP-binding motifs as well as two predicted transmembrane domains. Interestingly, the atlastin-3 gene maps to an area within the SPG17 (Silver syndrome; MIM 270685) locus (33) and thus may be a candidate gene for SPG17. The atlastin-2 gene is in the vicinity of the SPG4 gene spastin on chromosome 2.
The transmembrane structure we propose for atlastin-1 (Fig.  7C) is not typical of GBPs, the family of large GTPases to which the atlastins are most homologous (11,12,30). However, although the atlastins and GBPs all share an RD loop instead of (N/T)KXD in the third motif within the classical guanylatebinding triad, there are several structural differences between the atlastin and GBP families (12,13,32). First, unlike atlastins, GBPs have an additional C-terminal ␣-helical domain that folds back to interact with more proximal areas (12). Second, many GBPs have a C-terminal CAAX (where A represents an aliphatic amino acid) motif for isoprenylation, which all of the atlastins lack (Fig. 1). Third, our study has demonstrated that atlastin-1 and likely atlastin-2 and atlastin-3 are integral membrane proteins with two putative transmembrane domains; by contrast, GBPs are not integral membrane proteins. In this regard, the atlastins are more reminiscent of the Fzo/mitofusins, large GTPases that span the outer mitochondrial membrane twice, with both N and C termini facing the cytoplasm (34,35). Because the Fzo/mitofusins are critical for proper mitochondrial fusion, atlastin-1 may also have roles in homo-or heterotypic fusion events. Finally, like many members of the dynamin/Mx/GBP superfamily (12,30,31,36), atlastin-1 forms homo-oligomers. Although our study indicates that atlastin-1 most likely assembles as a homotetramer, it is not clear whether it can also form heteromeric complexes with atlastin-2 and atlastin-3. It is also unclear whether the atlastins are able to form higher order structures, as has been shown for other members of the dynamin/Mx/GBP superfamily (12,31,36).
Although the functions of the atlastin proteins are unknown, several groups have identified atlastin-interacting proteins by yeast two-hybrid screening (32,37). Atlastin-1/human GBP3 has been shown to interact with human NIK/HGK, which is one of the group I germinal center kinases, a group of MAPK kinase kinase kinases that activates the SAPK/JNK cascade in a variety of cell types (32). The interaction with atlastin-1 occurs through the leucine-rich CHN domain of NIK/HGK; residues 1-129 of atlastin-1 appear to be sufficient for the interaction (32). Atlastin-2/ARL-6-interacting protein-2 has been shown to interact with ARL-6, an ADP-ribosylation factor-like protein that is predominantly cytosolic, but associates with membranes in response to GTP␥S (37). The C-terminal third of atlastin-2 (including the two proposed transmembrane domains) is sufficient for this interaction. Interestingly, several ADP-ribosylation factor and ADP-ribosylation factor-like proteins are important for recruiting proteins such as golgins and COPI to the Golgi apparatus (38,39). The functional implications of these identified atlastin-protein interactions are unknown, however.
How do mutations in the atlastin-1 gene alter the atlastin-1 protein? All but one of the reported atlastin-1 mutations in patients with SPG3A represent missense mutations; the other is a nucleotide insertion resulting in early termination in the C-terminal domain (13)(14)(15)(16). One missense mutation (R217Q) changes a highly conserved residue in the RD loop within the GTP-binding pocket of atlastin-1 and may thus alter GTP binding or GTPase activity (14). Although the other missense mutations occur at residues highly conserved among the atlastin family, their effects remain unclear. Interestingly, these mutations are located in both N-and C-terminal regions, suggesting that even minor changes in the primary sequence in multiple different regions may have profound structural or functional implications. These sequence changes could potentially interfere with oligomerization, proper membrane association, cellular localization, GTPase activity, and interactions of atlastin-1 with other proteins.
The atlastin-1 protein is most abundant in brain, although it is also present at much lower levels in other tissues, including lung, smooth muscle, adrenal gland, kidney, and testis. Within brain, atlastin-1 is prominently enriched in the lamina V pyramidal neurons in the cerebral cortex, a subpopulation of which exhibit a distal axonopathy in patients with SPG3A. These upper motor neurons project to lower motor neurons in the lumbar spinal cord, and their dysfunction results in a spastic paraparesis, the cardinal feature of HSPs. Because these neurons have among the longest axons in the central nervous system, their dysfunction in SPG3A patients may reflect a critical role for proper atlastin-1 function in this subpopulation of "long axon" upper motor neurons. Based on the subcellular localization of atlastin-1 to the Golgi apparatus and its structural similarity to members of the dynamin/Mx/GBP superfamily of GTPases, atlastin-1 may be important for proper Golgi membrane dynamics or vesicle trafficking.
How might defective Golgi membrane structure or vesicle trafficking cause a distal axonopathy in upper motor neurons with long axons? Defective transport along axons has been directly implicated in a number of hereditary long axonopathies, including the HSP SPG10 (mutations in the neuronal kinesin KIF5A) (8) and the hereditary neuropathy Charcot-Marie-Tooth type 2A (mutations in the KIF1B␤ motor protein) (40), as well as in an autosomal dominant form of lower motor neuron disease (mutations in p150 subunit of dynactin) (41). Although these motor proteins directly affect transport, proper formation of intracellular cargoes is important as well. Given the early onset of SPG3A and suggestions that it may be due to abnormalities of neuronal development (3,4), it is intriguing that treatment of hippocampal neurons with brefeldin A, a fungal metabolite that disrupts the Golgi apparatus, inhibits axonal growth (42). Thus, an intact Golgi apparatus is required for axonal growth (42), which may be particularly relevant for long axon formation during development of the central nervous system. Conceivably, mutations of atlastin-1 in patients with SPG3A function may result in impaired Golgi structure or function, leading to impaired axonal growth. Future studies of the effects of SPG3A patient mutations on Golgi structure and function as well as axonal transport and growth may clarify the cellular pathogenesis of the HSP SPG3A.