The X-ray Crystal Structure and Putative Ligand-derived Peptide Binding Properties of γ-Aminobutyric Acid Receptor Type A Receptor-associated Protein*

The γ-aminobutyric acid receptor type A (GABAA) receptor-associated protein (GABARAP) has been reported to mediate the interaction between the GABAAreceptor and microtubules. We present the three-dimensional structure of GABARAP obtained by x-ray diffraction at 1.75 Å resolution. The structure was determined by molecular replacement using the structure of the homologous protein GATE-16. NMR spectroscopy of isotope-labeled GABARAP showed the structure in solution to be compatible with the overall fold but showed evidence of conformation heterogeneity that is not apparent in the crystal structure. We assessed the binding of GABARAP to peptides derived from reported binding partner proteins, including the M3-M4 loop of the γ2 subunit of the GABAAreceptor and the acidic carboxyl-terminal tails of human α- and β-tubulin. There is a small area of concentrated positive charge on one surface of GABARAP, which we found interacts weakly with all peptides tested, but we found no evidence for specific binding to the proposed physiological target peptides. These results are compatible with a more general role in membrane targeting and transportation for the GABARAP family of proteins.

The ␥-aminobutyric acid receptor type A (GABA A ) 1 is a ligand-gated chloride ion channel that mediates inhibitory neurotransmission (1). The ␥-aminobutyric acid receptor type Aassociated protein (GABARAP) was identified as interacting with the M3-M4 loop of the ␥2 subunit of the GABA A receptor by yeast two-hybrid analysis and co-immunoprecipitation (2). The initial two-hybrid clone identified consisted of the carboxyl-terminal portion (residues 36 -117) of the complete 117-residue GABARAP protein. A distant homology to microtubule-associated protein 1A and microtubule-associated protein 1A and 1B light chain 3 (MAP-LC3) (3) identified GABARAP as a possible tubulin-binding protein. GABARAP was found to be associated with microtubules (2,4), and the highly basic aminoterminal region of the GABARAP protein was putatively identified as the tubulin binding region. Thus, it has been suggested that GABARAP serves to link the GABA A receptor (through interactions with the carboxyl-terminal 35-117 region) to the cytoskeleton (via the amino-terminal region) and is involved in organizing GABA A receptors at the synapse (2,5). Other experiments have not supported location of GABARAP at the synapse (6) but instead indicate interaction with gephyrin, a protein which itself binds tubulin and the glycine receptor (7) and has a role in the postsynaptic localization of GABA A receptors.
Sequence analysis (4, 6) identified a number of orthologues of GABARAP. In addition to MAP-LC3, these included yeast autophagy protein Aut7p (sequences from three species exist) and a protein involved in intra-Golgi transport (GATE- 16), also known as ganglioside expression factor 2. More recently, an additional homologue has been identified as an early estrogenregulated gene, gec1 (8). Three GABARAP homologues have been identified from cDNA data base searching (9); two are identical to ganglioside expression factor 2 and Gec1, and the third is very similar to Gec1.
Yeast Aut7p (also known as Apg8) has recently been shown to undergo lipidation by a phosphatidylethanolamine on the carboxyl-terminal glycine carboxyl group of the mature protein (10). The terminal arginine is removed during the maturation of the protein by a protease and the lipid transferred by proteins related in sequence to the ubiquitin ligase enzyme families E1 and E2. The lipidation is essential for membrane dynamics during autophagy. GABARAP, GATE-16, and MAP-LC3 all covalently attach to the human homologue of the E1 enzyme Apg7p (11). The recently determined crystal structure of GATE-16 (12) shows a ubiquitin-like fold with a unique amino-terminal helical extension. Both GABARAP and GATE-16 bind to ULK1, an unk51-like kinase involved in neurite extension, via a proline-serine-rich region (13).
We report here the crystal structure of GABARAP and use NMR to assess the reported interactions with partner proteins using short peptides derived from GABA A ␥2L and tubulin. The three-dimensional structure of GABARAP closely resembles that of GATE-16. Through peptide titrations monitored by heteronuclear NMR spectroscopy, we identify an apparently general low-affinity peptide-binding surface on GABARAP. However, under the conditions of these experiments, GABARAP fails to display significant specificity for peptides from either the carboxyl terminus of tubulin or the GABA A ␥2L M3-M4 loop sequence that was used as bait in the original yeast twohybrid screen. We discuss the implications of these results for the definition of the functional role of the GABARAP protein.

EXPERIMENTAL PROCEDURES
Expression and Purification-For crystallography and later NMR samples, the GABARAP coding region was directionally cloned into a nonfusion derivative of expression vector pET28a (Novagen Inc., Stamford, CT), and the resulting construct was transformed into Escherichia coli BL21(DE3) cells grown at 37°C in Luria Broth containing 30 mg/liter kanamycin. Protein expression was induced by the addition of 0.5 mM isopropyl-␤-D-thiogalactopyranoside when the cell culture reached an optical density of A 600 ϭ 0.6. Cells were harvested by centrifugation after 16 h. The protein was purified at 4°C by anion exchange (fast flow S-Sepharose; Amersham Biosciences, Inc.) on a NaCl gradient in 10 mm Tris (pH 7.0) eluting at about 0.4 M NaCl and gel filtration in 10 mM Tris, 0.5 M NaCl (pH 7.0) (S200 Sephadex; Amersham Biosciences, Inc.). The experimental relative molecular mass for GABARAP determined by electrospray ionization mass spec-trometry was 13,914.6 daltons. Earlier NMR samples were made using a pET-15b His-Tag construct. This was grown under essentially the same conditions but with ampicillin (100 mg/liter) replacing the kanamycin. The protein was purified on Ni-NTA resin and eluted with 0.5 M imidazole, concentrated, and further purified by gel filtration (S200 Sephadex; Amersham Biosciences, Inc.).
Crystallization and Data Collection-Recombinant GABARAP was crystallized at 22°C using the hanging drop vapor diffusion technique by mixing 2 l of protein at a concentration of 36 mg/ml in 10 mM Tris, 0.5 M NaCl (pH 7.0), and 2 l of reservoir solution containing 0.1 M Tris (pH 8.5), 20% (w/v) polyethylane glycol monomethylether 2000, and 10 mM nickel(II) chloride hexahydrate. Before flash-cooling, crystals were gradually equilibrated against reservoir solution containing, in addition, 20% polyethylene glycol 400. The final data were collected at 100K on station 14.1 at Council for the Cinwal Laboratory of the Research Councils Daresbury Laboratory using 1.488 Å monochromatic x-rays.
Structure Solution-Data were processed with MOSFLM/SCALA software (14 -16). The crystals were in space group P2 1 2 1 2 1 with a unit cell of a ϭ 30.16 Å, b ϭ 54.95 Å, and c ϭ 64.67 Å. These parameters predicted one molecule of GABARAP in the asymmetric unit with a FIG. 1. a, stereo representation of the superposition of the GABARAP and GATE-16 main chains. b, cartoon diagram of the GABARAP three-dimensional backbone structure. Region 1-35 is in green, region 35-68 is in red, and region 68 -117 is in blue. Therefore, the baits used in the two hybrids are the 35-117 region (colored in red and blue) and the 1-68 region (colored in green and red). c, backbone trace of GABARAP in the same orientation as b, showing residues discussed in the text. Conserved residues are shown in red for acidic residues, blue for basic residues, magenta for polar residues, green for residues involved in turns, and black for hydrophobic core residues. The nonconserved Phe 3 is shown in light gray, and water molecules involved in the region of the salt bridges are shown in cyan. This figure was prepared with MOLSCRIPT (32) and RASTER3D (33). solvent content of 30%. The coordinates of the GATE-16 crystal structure (Protein Data Bank accession code 1EO6; chain A) (12) were used for the rotation and translation searches in MOLREP (17) and yielded a solution with an initial R crys of 0.512 (against a previously obtained lower resolution dataset). Refinement against 1.75 Å data was performed using CNS (18) with rounds of manual rebuilding in O (19). Final refinement was performed using Refmac5 (20) with TLS (21) parameters for GABARAP as a single body. Due to the wavelength used being close to the absorption edge of the nickel atom in the crystal, anomalous refinement was used in CNS with calculated anomalous scattering factors. In Refmac5, this phenomenon was approximated by altering the constant term of the Ni 2ϩ scattering factor to Ϫ7.00. With either program, the greatest effect was on the B factor of the nickel atom, which was reduced to a value similar to the surrounding atoms; the effect on the R-factor was small.
NMR Procedures-A nearly complete set of backbone resonance assignments for double 15 N, 13 C isotope-labeled GABARAP has been obtained using triple resonance spectroscopy and is reported elsewhere by us (22) and independently (23). Titrations of candidate peptide ligands with GABARAP were performed in the following manner. 1-5-l aliquots of up to 50 l of an approximately 20 mM solution of the peptide were added to 0.5 ml of a 0.3 mM solution of uniformly 15 N-labeled GABARAP at 20°C. A two-dimensional 15 N, 1 H HSQC spectrum was recorded at each titration point, utilizing pulse field gradient sensitivity enhancement with water flip-back pulses to minimize cross-saturation (24). The sample pH was verified after each addition of the peptide, and control pH titration experiments were performed to check the specificity of any cross-peak perturbations. Peptides were synthesized by commercial suppliers, and the composition and purity (Ͼ95%) of each were verified by high performance liquid chromatography and mass spectrometry.

RESULTS AND DISCUSSION
Crystal Structure and Comparison with GATE-16 -The crystal structure of GABARAP was solved at 1.75 Å resolution using molecular replacement with chain A of GATE-16 (56% sequence identity; Protein Data Bank accession code 1EO6) (12) as the starting model (Fig. 1a). The final refined GABARAP structure has an R-factor of 0.203 and a free Rfactor of 0.230 (Table I). All 117 amino acids are fully modeled along with a nickel cation, which forms a crystal contact, and 100 water molecules. The nickel atom is octahedrally coordinated by four clearly resolved water molecules and the side chain N⑀ atoms of His 69 in one GABARAP molecule and His 99 in a symmetry-related molecule.
The GABARAP structure consists of a four-stranded ␤-sheet, with two ␣-helices adjacent to each face (Fig. 1b). The major portion of the structure, from residue 28 onward comprising the four ␤ strands and the last two ␣-helices, closely resembles the ubiquitin fold found in a large number of proteins. These include not only ubiquitin and homologues, but the ras binding domain family, the first subdomain of the FERM fold, and 2Fe-S ferrodoxin related domain. Against this similarity, the two ␣-helices at the amino terminus of GABARAP and GATE-16 are a unique feature of the protein fold.
There are few differences between the GABARAP structure and that of GATE-16 (Fig. 1a). The conformations of the carboxyl terminus (residue 111 onward) are different in the three structures (GABARAP and two GATE-16 conformers), possibly reflecting plasticity in this region that is affected by different crystal contacts in each case. Excluding the carboxyl terminus, the root mean square deviation for the GABARAP C␣ atoms is 0.97 Å and 0.84 Å from chain A and chain B of GATE-16, respectively: the root mean square deviation between residues 1-111 (C␣ atoms) of the two chains of GATE-16 is 0.50 Å. The regions of greatest divergence occur at the end of the second helix (residues 18 -25), the loop (residues 37-40) between the first and the second ␤-strands, and the loops at the beginning and end of the third ␤-strand (residues 69 -74 and 81-87).
The conformation of the first loop differs due to the presence of a proline-lysine dipeptide at residue positions 37-38 in the GABARAP sequence, where there is serine-glycine in GATE-16. Glycine 42 in the GABARAP sequence allows the subsequent strand to adopt the same conformation as GATE-16. Similarly, the difference in the 69 -74 region is most marked at residue 71, where there is an arginine residue in GABARAP and a proline in GATE-16. In the other two regions, the variation in the two GATE-16 structures is nearly as great as the variation of GABARAP from GATE-16.
Sequence Conservation in the GABARAP Family-To evaluate the conservation patterns in the GABARAP family, a BLAST search of the human GABARAP sequence was run against the SwissProt data base including the translated EMBL data base (June 2001). SwissProt contains a large number of GABARAP-related proteins from Arabidopsis (8), as well as from Caenorhabditis elegans (3) and Drosophila (3). These homologous proteins are in addition to the sequences discussed in the introduction. 11 loci encoding intact sequences of GABARAP/MAP-LC3-like proteins are found in the Ensembl annotation (www.ensembl.org) of the human data base. These loci correspond to GABARAP and the three homologues reported in Ref. 9, a new sequence most closely related to GABARAP (GABARAPRL4) on chromosome 1q32 (although this lacks the conserved glycine that is reportedly lipidated in some family members), genes encoding the MAP-LC3 found on chromosome 12 and 16, two genes corresponding to deposited cDNA homologues of MAP-LC3, and two additional MAP-LC3like sequences on chromosomes 9 and 12. These sequences are shown as a sequence alignment in Fig. 2.
The residues totally conserved across all sequences in this family fall into several groups (Fig. 1c). There are charged residues that form hydrogen bonds and salt bridges between secondary structure elements (Lys 6 to Asp/Glu 100 , Arg 14 to Asp 102 , Glu 34 to main chain N of residues 4 and 5, Asp/Asn 45 to Arg 67 , Asn/Asp 81 to Asp/Glu 100 , and Asp 102 to Tyr 106 ). The waters with the lowest B factors are also found in the region of these salt bridges, where they mediate extra hydrogen bonding (Z12, 86, 89, 90, 93 and 94 are shown in Fig. 1c). There are conserved hydrophobic core residues (Pro 30 , Leu 50 , Val 51 , Ile 64 , Leu 70 , Val 80 , Tyr 95 , Phe 104 , and Tyr 106 ). Tyr 106 is particularly important to this family because it forms a hydrogen bond to Asp 102 and provides an aromatic ring for one of the aromatic side chains near the amino terminus to pack onto (Phe 3 in GABARAP and Trp 3 in GATE-16). There are a series of residues important for turns and initiating secondary structure: Pro 26 (loop that joins amino-terminal helices to the core ubiq- uitin fold), Pro 52 (end of ␤-strand), Ala 75 (small side chains prevent clashes with residue 111 at the start of the ␤-strand), and Ser 88 (turn leading into the carboxyl terminal ␣-helix). NMR Spectroscopy of GABARAP-We have also examined the structural and dynamic properties of GABARAP using multidimensional heteronuclear NMR spectroscopy. A nearly complete set of backbone resonance assignments for double 15 N, 13 C isotope-labeled GABARAP has been obtained using triple resonance spectroscopy and is reported elsewhere by us (22) and by others (23). In summary, the patterns of chemical shifts, interproton nuclear Overhauser effects, and backbone scalar coupling constants are consistent with the secondary structure elements revealed by the x-ray structure of GABARAP.
An interesting aspect of the NMR spectrum of GABARAP is that it contains features that indicate conformational heterogeneity of the amino-terminal region. Namely, we observed two amide NH cross-peaks in two-dimensional 15 N, 1 H-correlated spectra for residues Glu 17 , Glu 19 , Lys 20 , and Leu 105 (and two others that could not be assigned), indicating two-state conformational exchange on a slow timescale. In addition, it has proved impossible to unambiguously locate a number of resonances from neighboring regions of the structure (corresponding to Met 1 -His 9 , Phe 11 -Arg 15 , Arg 22 , Lys 47 , Glu 100 , Phe 103 , and Phe 104 ). It is possible that, for these signals, the conformational exchange is occurring on a timescale that leads to strong line-broadening and low-intensity NMR cross-peaks. For GABARAP, the residues that exhibit evidence of conformational exchange cluster together around the amino-terminal helices (helices-1 and -2) and the internal salt bridges connecting Arg 14 and Lys 6 with Asp 102 and Glu 100 , respectively. The precise nature of the conformational exchange that gives rise to these spectral characteristics is unclear at present. The GABARAP crystal structure exhibits no clear-cut evidence for conformational heterogeneity and therefore does not assist in rationalizing the solution NMR observations. It is worth noting that the NMR characteristics indicating conformational exchange in GABARAP persist for different values of the sample temperature (10°C to 30°C) and protein concentration (0.3-2.0 mM) and remain in samples containing up to 4 M urea denaturant (at pH 7.0).
Nitrogen-15 relaxation analysis (26,27) of GABARAP yields a typical pattern of amide NH bond generalized order parameters (S 2 ), with high values (Ͼ0.7) in the secondary structure elements and lower values in the loop regions (data not shown). The Lipari-Szabo formalism-derived rotational correlation time (28) is rather high for a protein of this nominal molecular mass (ϳ16,000 daltons; c ϳ 9.7 ns at 30°C). However, we have detected no evidence, either by NMR or other methods, that GABARAP has any tendency to form transient oligomers. The a The peptide sequences are listed using the one-letter amino acid code. Peptide VI was blocked by carboxyl-terminal amidation. Irrelevant indicates that the sequence of the peptide is derived from molecules with no anticipated physiological involvement in GABARAP function. b Residues with cross-peaks shifted by shifted by Ͼ 1 linewidth are denoted in bold and underlined text. Residues with cross-peaks shifts on the order of 1 linewidth are shown in bold, and those displaying smaller but nevertheless significant changes are shown in plain text. dynamic properties of GABARAP are being further investigated in our laboratory.
GABARAP-Peptide Binding Probed by NMR-The interaction of GABARAP with short peptides derived from the GABA A receptor ␥2L subunit (the predicted interaction site) and with the acidic carboxyl-terminal tails of ␣and ␤-tubulin, which are regarded as likely microtubule interaction sites for GABARAP, has been probed using two-dimensional 15 N, 1 H HSQC NMR spectroscopy. The peptides used in the NMR ligand titrations are detailed in Table II, along with a summary of the spectral perturbations observed in each case.
Peptide II corresponds to the 18-residue segment of the GABA A receptor subunit ␥2L that is reported to combine with GABARAP-(36 -117) to activate reporter gene transcription in a yeast two-hybrid interaction test (2). Peptide I is a truncated form of peptide II that corresponds to a 13-residue segment of ␥2L that fails to interact with GABARAP-(36 -117) in a similar assay (2). Peptides III and IV correspond to the carboxyl-terminal tails of the ␣and ␤-subunits of tubulin, regions that are thought to be exposed to solvent in intact microtubules (29). Peptides V and VI, which were used as controls, are derived from two proteins with no reported involvement in binding to GABARAP and were randomly selected from a number of "irrelevant" peptides available in our laboratory. Fig. 3 shows the results of the GABARAP-peptide III titration in the form of superposition of a region of the NMR spectra obtained for a number of additions of peptide. For each titration performed, the peptide effected selective perturbations of a number of GABARAP cross-peaks, indicating structurally localized interaction of the peptide with the GABARAP protein.
Control experiments were performed to check that the chemical shift changes are not attributable to small changes in the buffer conditions (e.g. pH, protein concentration) arising from addition of the peptide solution. The fact that the cross-peaks shift in a linear fashion across the spectrum, without diminution in intensity, indicates single-site peptide binding with fast exchange kinetics. Peptides I-VI were titrated to ϳ3ϳ5-fold excess over the GABARAP protein (concentration, ϳ0.5 mM). Qualitatively, the magnitude of the chemical shift perturbations over the whole titration is in the following order: peptide III ϳ peptide IV ϳ peptide V Ͼ peptide II Ͼ peptide I ϳ peptide VI. Only in the case of peptides III and IV was there evidence that the chemical shift perturbation was tailing off, indicating approach to saturation. Quantitative analysis of the change in chemical shift with peptide concentration yielded an estimate of the dissociation constant for peptides III and IV in the region of 0.1-0.2 mM. Dissociation constants for the other peptides were too high to be analyzed accurately. The data suggest that peptides I, II, and VI bind extremely weakly to GABARAP.
The pattern of residues that exhibited peptide-dependent chemical shift changes is remarkably consistent across the whole series of peptides. Fig. 4, AϪE, shows a three-dimensional representation of the GABARAP structure, highlighting the positions of the atoms with peptide-perturbed chemical shifts. In each case, there appears to be a bias of the larger chemical shift effects that is toward the surface of the protein opposite to the extreme amino and carboxyl termini, with clustering that appears to align with the exposed edge of ␤-strand 2 (e.g. residues Tyr 49 , Phe 50 , and Val 51 ) and helix-3 (e.g. Phe 60 and Ile 64 ). Notably, none of the peptides alters the peak-doubled characteristic of those resonances from the amino terminus that indicate conformational heterogeneity. It seems the binding of peptides is not so strong as to perturb the global conformation of GABARAP in even a minor way.
The NMR peptide titration experiments, designed to probe the interaction of putative protein ligands with GABARAP, yielded results that were consistent with relatively weak interactions (K d Ͼ ϳ0.1 mM). Whereas structurally specific crosspeak perturbations were observed for peptides with sequences derived from both the GABA A receptor ␥2L subunit and the carboxyl-terminal tails of ␣and ␤-tubulin, similar effects were also observed for a randomly selected peptide from an irrelevant context (phosphopeptide V). A notable feature of these results is that both peptide II, which is overall positively charged, and peptides III and IV, which are strongly negatively charged, effected chemical shift changes in the GABARAP spectrum consistent with a similar binding site location (Fig. 4,  BϪD). The protein surface electrostatic properties of GABARAP indicate that the binding is to a positively charged area with a surface pit in the vicinity of ␤-strand 2 (Fig. 4F). There is a single notable uncharged surface region focused  Table II: large sphere, Ͼ1 linewidth; medium sphere, approximately 1 linewidth; and small sphere, Ͻ1 linewidth. F, space-filled representation of the GABARAP structure colored according to the surface electrostatic potential computed with the program GRASP (25), with orientation identical to that in AϪE. around residues Pro 84 and Pro 85 , but this does not appear to be a component of the ligand binding site.
Overall, the NMR experiments have revealed that fulllength GABARAP protein has a weak and nonspecific affinity for peptides that appears rather insensitive to the chemical composition. Apart from the apparent promiscuity of GABARAP for weak interactions with peptides of different composition, rationalization of these observations with the reported ability of GABARAP to bind the GABA A receptor ␥2L subunit directly with sufficient strength to be detectable by in vitro pull-down assay (2) is not facile. GABARAP was originally identified in a yeast two-hybrid screen for molecules that interact with the M3-M4 loop of the ␥2L subunit of GABA A receptors. Notably, only a partial cDNA corresponding to GABARAP residues 36 -117 was originally isolated from the library screen. Full-length GABARAP was ineffective at ␥2L binding in the yeast two-hybrid assay, but it was shown that all of GST-GABARAP-(36 -117), GST-GABARAP-(1-68), and GST-GABARAP-(1-117) were able to bind ␥2L in vitro. These results implicate GABARAP residues 36 -68 in ␥2L binding, and this observation could be considered broadly in line with our NMR observations of ␥2L peptide binding (albeit with very low affinity) in the region of residues 49 -51 (␤-strand 2). However, given the compact three-dimensional structure of GABARAP, it is difficult to conceive of the GABARAP-(37-117) and GABARAP-(1-68) deletion constructs possessing a stable globular fold (Fig. 1b). Scission at residue 35 would remove the amino-terminal helices-1 and -2 and the second ␤-strand of the four-stranded ␤-sheet. Cleavage at residue 68 splits the molecule in two halves and deletes two ␤-strands and helices-3 and -4. The yeast two-hybrid screen results suggest that an 18-residue fragment of ␥2L (which is too small to be structurally ordered) can interact with the GABARAP-(36 -117) deletion mutant, a fragment that we must anticipate is not folded in the manner detected for the intact protein. These considerations suggest that due care should be adopted in interpreting the structure-function relationships underlying the results of experiments performed with GABARAP-(36 -117) deletion constructs, particularly if invoking the folded form of the intact protein as part of the rationalization for biochemical interactions.
GABARAP has also been reported to bind microtubules, and there has been speculation that the basic amino-terminal segment of the protein (corresponding to helices-1 and -2 in the x-ray structure) represents a site of interaction for acidic sites on ␣and ␤-tubulin (2). For example, omission of the first 35 amino acids of GABARAP ablates microtubule binding (2). In addition, a peptide corresponding to the first 22 amino acids of GABARAP promotes microtubule assembly; a scrambled sequence peptide with the same amino acid composition lacks this activity (4). However in NMR titration experiments with peptides corresponding to the flexible tails of both ␣and ␤-tubulin proteins, we find that they do interact (albeit weakly), but with an interaction site on a part of the surface that is on the opposite side of the GABARAP molecule, away from the aminoterminal helices. The electrostatic properties of GABARAP (Fig. 4F) demonstrate that the density of basic residues in the primary sequence of the amino-terminal region does not translate into a particularly positively charged surface in this region of the three-dimensional structure. Presumably, this effect occurs as a result of the neutralization of some of these residues in (partially) buried salt bridge interactions (e.g. Lys 6 to Glu 100 and Arg 14 to Asp 102 ).
Confocal microscopy experiments have also shown that GABARAP does not co-localize exclusively with either the GABA A receptor or gephyrin (a second protein thought to be involved in organizing GABA A receptor at the postsynaptic membrane) and tends to be associated with endoplasmic reticulum or Golgi membranes (6). GATE-16 binds to N-ethylmaleimide-sensitive factor, a protein that uses ATP to disassociate protein complexes that ensure correct membrane fusion. GABARAP has been shown to bind to N-ethylmaleimide-sensitive factor also (30). It now seems likely that the GABARAP/ GATE-16 family and probably the microtubule-associated protein 1A LC3 family as well are involved in the process of protein transport through the Golgi. The 10 different human gene sequences identified in this family of proteins give an opportunity for there to be specificity in interactions between GABARAP family members and the proteins they assist in transport. Alternatively, different GABARAP family members may act on different membranes as proposed for the rab family of G proteins (31). However, there is also likely to be some redundancy in the function of these proteins, which will make identification of the specific roles of GABARAP family members a challenge.