Solution Structure of Microtubule-associated Protein Light Chain 3 and Identification of Its Functional Subdomains*

Microtubule-associated protein (MAP) light chain 3 (LC3) is a human homologue of yeast Apg8/Aut7/Cvt5 (Atg8), which is essential for autophagy. MAP-LC3 is cleaved by a cysteine protease to produce LC3-I, which is located in cytosolic fraction. LC3-I, in turn, is con-verted to LC3-II through the actions of E1- and E2-like enzymes. LC3-II is covalently attached to phosphati-dylethanolamine on its C terminus, and it binds tightly to autophagosome membranes. We determined the solution structure of LC3-I and found that it is divided into N- and C-terminal subdomains. Additional analysis using a photochemically induced dynamic nuclear polarization technique also showed that the N-terminal subdomain of LC3-I makes contact with the surface of the C-terminal subdomain and that LC3-I adopts a single compact conformation in solution. Moreover,

are abundant in the nervous system and are major components of the neuronal cytoskeleton (2,3). They share common distributions and physiological functions. For example, both are prominently expressed throughout the brain and associate with microtubules via a microtubule-binding domain found in the N-terminal region of MAPs. This domain of MAP1B was identified as a basic region containing KKE(E/I/V) repeats (4) and may interact with the C-terminal acidic region of tubulin. MAP1A and MAP1B act cooperatively with their light chains, MAP-LC2 and MAP-LC1, respectively. Complexes of MAP and its corresponding light chain are synthesized as a single large polyprotein and then are post-translationally cleaved into two distinct chains (5,6). In contrast, MAP-LC3 is not synthesized from either MAP1A/LC2 or MAP1B/LC1 polyprotein mRNAs. However, MAP-LC3 co-immunoprecipitates with either MAP1A or MAP1B, indicating that this light chain is a common partner of both MAPs (7). On the other hand, the expressions of MAP1A and MAP1B are known to differ as neurons grow. MAP1B is highly expressed in developing neurons, whereas the expression level of MAP1A is highest in mature neurons; in other words, MAP1A and MAP1B display complementary patterns of expression (7). Although the binding activities and the co-localization with MAPs suggest that MAP-LCs are related to the development and the differentiation of neurons, the detailed physiological functions of MAPs and MAP-LCs are not yet understood.
In yeast, autophagy is induced by nutrient starvation to provide the minimal nutrients for cell maintenance. To date, at least 16 genes required for autophagy have been identified, and their products are referred to as the Apg/Aut/Cvt (Atg) family (8 -10). MAP-LC3 is a mammalian homologue of Apg8/Aut7/ Cvt5 (Atg8), which is essential for the formation of autophagosomes and cytoplasm-to-vacuole targeting (Cvt) vesicles (11,12). After synthesis, Atg8 is cleaved by Apg4/Aut2 (Atg4), a yeast cysteine protease, exposing the C-terminal glycine to solvent (13,14). Subsequently, the cleaved Atg8 is modified by Apg7/Cvt2 (Atg7) and Apg3/Aut1 (Atg3), which act as E1 and E2 enzymes of the ubiquitination system, respectively (15)(16)(17)(18)(19)(20). The C-terminal glycine of Atg8 is covalently attached to phosphatidylethanolamine (PE). The PE-modified form is tightly bound to autophagosomes, whereas the unmodified form is localized in the cytosolic fraction (13,19). In addition, immunoelectron microscopy has shown that Atg8 is localized in the precursor and mature structure of autophagosomes (11). These findings suggest that Atg8 plays a key role in the formation of autophagosomes, which agrees with the finding that a lack of Atg8 or the inhibition of its modification causes a defect in autophagy (13). Interestingly, MAP-LC3 is modified by PE in the same ways as Atg8 (14,(21)(22)(23)(24). Unmodified and modified forms of MAP-LC3 are referred to as LC3-I and LC3-II, respectively. Taken together, the data suggest that MAP-LC3 is involved in the formation and/or the intracellular trafficking of autophagosomes in mammalian cells.
Two additional proteins have been identified as human homologues of Atg8: GABA A receptor-associated protein (GABARAP) (25) and Golgi-associated ATPase enhancer of 16 kDa (GATE-16) (26). Structural analyses of GABARAP and GATE-16 using x-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy have already been reported (27)(28)(29)(30)(31)(32)(33)(34). These reports indicate that the tertiary structures of GABARAP and GATE-16 are very similar. The structures include an N-terminal subdomain composed of two ␣-helices and a C-terminal subdomain that adopts a ubiquitin fold. Interestingly, Coyle et al. (31) demonstrated that GABARAP undergoes a conformational change in the N-terminal region, which may be involved in the polymerization of tubulin. NMR experiments have shown that some of the heteronuclear single-quantum correlation (HSQC) signals for the amino acids in the N-terminal subdomain indicate NH cross-peak doubling (28,29,34), which also supports a slow conformational change in the Nterminal subdomain of GABARAP.
In the present study, we did not observe NMR signals with peak doubling in the spectra of LC3-I, and we were able to use NMR spectroscopy to successfully determine the solution structure of human LC3-I. In addition, we carried out a photochemically induced dynamic nuclear polarization (photo-CIDNP) experiment to obtain additional information about the N-terminal conformation of LC3-I. The photo-CIDNP approach is based on the cyclic photochemical reactions between a photoexcited dye and the amino acid residues located on the surface of a protein (35)(36)(37). This technique selectively detects the aromatic amino acids His, Trp, and Tyr when these side chains are accessible to the dye. By investigating whether tyrosine residues in LC3-I are located on the surface of the protein, we obtained evidence that LC3-I adopts a single conformation in solution. Moreover, we explored the physiological significance of the two subdomains in LC3-I by interaction assays using surface plasmon resonance and HSQC cross-peak perturbation. We propose here that the two subdomains play different roles in the function of MAP-LC3, suggesting that MAP-LC3 acts as an adaptor protein in the autophagy system.

Expression and Purification of LC3-I and Its Mutant-Recombinant
LC3-I was expressed in Escherichia coli strain BL21 as a glutathione S-transferase fusion protein. Uniformly 15 N-labeled and 13 C/ 15 N-labeled LC3-I proteins were obtained by growing E. coli cells in minimal medium containing 15 NH 4 Cl or a combination of [ 13 C]glucose and 15 NH 4 Cl, respectively. The cells were harvested by centrifugation and suspended in 25 mM sodium phosphate (pH 7.5) containing 150 mM NaCl. The cell lysate was then centrifuged at 10,000 ϫ g for 10 min, and the supernatant was applied to an immobilized glutathione column (Amersham Biosciences). Bound protein was treated with PreScission protease (Amersham Biosciences) to remove glutathione S-transferase. Finally, LC3-I protein was purified using a BioScale-S2 cation exchange column (Bio-Rad) in 50 mM Tris (pH 8.0). The protein was eluted with an NaCl gradient up to 500 mM. The expression plasmid for the LC3-I mutant (LC3-I⌬N) lacking the N-terminal subdomain (residues 1 to 29) was generated using the plasmid containing the full-length LC3-I as a template. LC3-I⌬N protein was prepared according to the same protocol as that used for LC3-I. The purities of LC3-I and LC3-I⌬N were confirmed by SDS-PAGE and mass spectrometry.
NMR Spectroscopy-All NMR samples contained 0.8 -1.0 mM non-, 15 N-, or 13 C/ 15 N-labeled LC3-I in 25 mM sodium phosphate (pH 7.0), 100 mM NaCl, 0.02 mM NaN 3 , and 10 or 100% D 2 O. The non-labeled LC3-I⌬N was also concentrated to 0.8 mM. Prior to the NMR experiments, centrifugation at 15,000 ϫ g was used to confirm that no aggregation was present in the samples. All of the NMR spectra were acquired at 298 K on Bruker DMX500 and AV800 spectrometers. The assignments of 1 H, 13 C, and 15 N resonances of the backbone and the side chains were performed using a series of three-dimensional NMR experiments (38). For the structure calculations of LC3-I, distance restraints were collected from 15 N-and 13 C-edited NOESY three-dimensional experiments and two-dimensional homonuclear NOESY with 85-ms mixing times. Dihedral angle () restraints obtained from a HNHA experiment were also used in the calculations. Two hours after dissolving LC3-I in 100% D 2 O, slowly exchanging amide protons were identified from the 1 H-15 N HSQC spectrum. The steady-state heteronuclear NOE values were collected from spectra recorded with and without 1 H saturation (39,40). Titration experiments were performed by the addition of dodecylphosphocholine (DPC) at final concentrations of 0.01 to 30 mM, and the 1 H-15 N HSQC spectrum was recorded at each titration point. The crosspeak perturbation (⌬␦) was quantified by the following equation: ⌬␦ ϭ [(5⌬ 1 H) 2 ϩ (⌬ 15 N) 2 ] 1/2 , where ⌬ 1 H and ⌬ 15 N are the changes in the chemical shifts along the 1 H and 15 N axes, respectively. The 1 H photo-CIDNP difference spectrum was recorded on a Bruker DMX500 with a continuous-wave argon ion laser (Spectra Physics) (35,41). Blue-green light (4 W) was introduced into the NMR tube through an optical fiber inside a coaxial Pyrex insert (Wilmad) dipped in the solution and was controlled by a mechanical shutter connected to the NMR spectrometer. The sample contained 0.8 mM LC3-I protein and 0.4 mM flavin mononucleotide as a photosensitizer in the NMR sample buffer. Light and dark free induction decays were alternately acquired with and without a 0.5-s light flash prior to a 90°RF pulse, respectively, and were integrated 16 times. The CIDNP difference spectrum was obtained by subtraction of the dark spectrum from the light spectrum. All NMR spectra were processed and analyzed using NMRPipe (42) and PIPP (43) software.
Structure Calculations-The solution structure of LC3-I was calculated with a distance geometry-simulated annealing protocol using X-PLOR software (44). The structure calculations and NOE peak assignments were performed iteratively and manually. NOE cross-peak intensities were classified as strong, medium, weak, and very weak, and assigned to restraints of 1.8 -2.8, 1.8 -3.5, 1.8 -5.0, and 1.8 -6.0 Å, respectively, with appropriate pseudo atom corrections (45). The hydrogen bonds were identified according to slow exchange of the amide proton with the solvent deuterium, and their partners were assigned on the basis of the early structure calculations. The hydrogen bond data were used as distance restraints of 2.3-3.3 and 1.3-2.3 Å for N-O and HN-O atom pairs, respectively, and were incorporated into the late stages of structure calculations. Finally, a simulated annealing protocol was applied using 12,000 steps at high temperature (1,000 K) and 8,000 steps for the cooling process. A total of 50 structures were calculated, and the 15 lowest energy structures were used for further calculation of the energy-minimized average structure. The quality of the obtained structure was analyzed with MOLMOL (46) and PROCHECK-NMR (47) software. The detailed experimental data and the structural statistics are summarized in Table I. The coordinates have been deposited under Protein Data Bank accession number 1V49.
Binding Assay by Surface Plasmon Resonance-The interactions of LC3-I and its mutant with purified tubulin (Cytoskeleton) were studied using a BIACORE J biosensor (Biacore). The tubulin solution was centrifuged (50,000 ϫ g) at 4°C for 10 min to remove aggregated protein. Immediately after centrifugation, the tubulin was immobilized on a CM5 biosensor chip using an amine-coupling reagent (Biacore), and uncoupled tubulin was washed out with 5 mM NaOH. Binding assays were performed at 25°C in 25 mM HEPES (pH 7.0). Prior to use, LC3-I and LC3-I⌬N solutions were passed through a 0.45-m filter. For each protein, 180 l of solutions at different concentrations (2.5 to 10 M) were injected at 60 l/min over immobilized tubulin and control flow cells. The control sensorgrams were subtracted from those obtained with tubulin. After each injection, regeneration was performed with 60 l of HEPES (pH 7.0) containing 400 mM NaCl.
Co-sedimentation Assay-Purified tubulin (Cytoskeleton) was dissolved in PEM buffer (100 mM PIPES, pH 6.6, 1 mM EGTA, and 1 mM MgCl 2 ) at a final concentration of 15 M. LC3-I, LC3-I⌬N, and tubulin solutions were centrifuged (50,000 ϫ g) at 4°C for 10 min, and their supernatants were used for experiments. For the polymerization of tubulin, GTP and paclitaxel (Sigma) were added to the tubulin solution at final concentrations of 1 mM and 50 M, respectively, and the solution was incubated at 37°C for 20 min. Following the incubation, the poly-merized tubulin was diluted to 7.5 M with PEM buffer. The LC3-I and LC3-I⌬N solutions were added to the tubulin solution at final concentrations of 5 M. After an additional incubation at 37°C for 10 min, centrifugation (50,000 ϫ g) was carried out at 37°C for 20 min. The pellets were resuspended in the original volume of PEM buffer and the centrifugation procedure was repeated. The final pellets were analyzed by SDS-PAGE followed by staining with Coomassie Brilliant Blue. As a negative control, 400 mM NaCl was added into the LC3-I mixture solution after the tubulin polymerization reaction and processed as above.

RESULTS
Structure Determination of LC3-I-To calculate the solution structure of LC3-I, a total of 2136 distance restraints and 55 dihedral restraints were obtained from the NMR spectra. The final simulated annealing calculation using these restraints gave 50 structures. Of these, the 15 structures with the lowest energy values were selected and used to calculate the energyminimized average structure. In the selected 15 structures, there were no distance restraints with a violation value larger than 0.20 Å and no dihedral angle restraints with a violation larger than 2°. The energy-minimized average structure did not include residues in the disallowable regions of the Ramachandran plot. The ensemble of the 15 structures are presented in Fig. 1A, and the backbone heavy atoms of the well defined regions gave a r.m.s. deviation value of 0.59 Ϯ 0.10 Å. The details of the structural statistics for LC3-I are shown in Table I.
The solution structure of LC3-I includes four ␣-helices (H1, H2, H3, and H4) and a central ␤-sheet composed of four strands (S1, S2, S3, and S4) (Fig. 1B). Strands S1 and S4 are parallel and are located in the middle of the ␤-sheet. Strands S2 and S3 are antiparallel to S1 and S4, respectively. The ␤-sheet bends around the helix H3. The global conformation of LC3-I is very similar to those of the other Atg8 mammalian homologues, GABARAP and GATE-16 (27, 30 -33). These structures are composed of a ubiquitin fold (i.e. helices H3 and H4 and a central ␤-sheet of LC3-I) and additional N-terminal ␣-helices (i.e. helices H1 and H2 of LC3-I), which we refer to as the Cand N-terminal subdomains, respectively (Fig. 1). The two subdomains are folded into a single globular conformation by hydrophobic cores. One of the cores is formed between the N-and C-terminal subdomains. The helix H2 protrudes from the Cterminal subdomain and is located along the surface of the central ␤-sheet. The helix H1 also covers the side chains of the residues, such as Ile-34 and Tyr-110 on the ␤-sheet (Fig. 1B). Thus, the two ␣-helices of the N-terminal subdomain cover the surface of the C-terminal subdomain and contribute to the formation of the hydrophobic environment. The NMR data indicated that the 1 H ␥11 and 1 H ␦1 resonances of Ile-34 showed an extreme upfield shift (Ϫ0.295 and Ϫ0.992 ppm, respectively) because of the close proximity of some aromatic rings derived from Phe-7, Phe-108, and Tyr-110. Another hydrophobic core is found in the C-terminal subdomain and is surrounded by helices H3 and H4 and the central ␤-sheet. Residues Ile-35 and Ile-67 are located on strand S1 and helix H3, respectively, and their side chains are directed toward the interior of the Cterminal subdomain (Fig. 1B). Like Ile-34 in the core between the two subdomains, the resonances of Ile-35 1 H ␦1 and Ile-67 1 H ␦1 are upfield-shifted (0.230 and 0.183 ppm, respectively) because of the ring current effects from the side chains of Phe-52, Phe-79, and Tyr-113 that participate in the core.
Additional Information about the N-terminal Conformation-According to the LC3-I structure calculations, helices H1 and H2 are located on the surface of the central ␤-sheet in the C-terminal subdomain. In addition, the steady-state heteronuclear NOE values of the residues included in helix H1 are relatively high (the average value of the heteronuclear NOEs for the residues Phe-7 to Arg-10 was 0.75 Ϯ 0.10) ( Fig. 2A), which indicates low flexibility in solution. However, the conformation of helix H1 is less defined in the overall structure of LC3-I (Fig. 1A). To obtain additional information about the spatial orientation of helix H1, we carried out a photo-CIDNP approach that detects only specific amino acid residues (His, Trp, and Tyr) that are located on the protein surface. LC3-I has three histidine and four tyrosine residues but lacks tryptophan. A tyrosine residue, Tyr-110, is located on the central ␤-sheet of LC3-I and its side chain makes contact with helix H1 in the solution structure (Fig. 1B). Assuming that the less defined structure of helix H1 is caused by the flexibility of the region, the side chain of Tyr-110 should be accessible to photoexcited dye. Accordingly, a CIDNP effect can be detected for Tyr-110. The signal derived from this tyrosine residue is a good indicator of whether or not helix H1 covers the surface of the central ␤-sheet.
The photo-CIDNP difference spectrum of LC3-I is shown in Fig. 2B. We observed five absorptive resonances and one weak emissive resonance. The five absorptive lines were assigned to the resonances of the three histidine residues (Fig. 2B), suggesting that the side chains of all three histidine residues (His-27, His-57, and His-86) are exposed to solvent. The emissive line in the CIDNP difference spectrum was assigned to the resonance derived from Tyr-38 (Fig. 2B), which indicates that the side chain of Tyr-38 is located on the surface. However, the side chains of other tyrosine residues (Tyr-99, Tyr-110, and Tyr-113) must be completely buried in the interior of the LC3-I structure because a CIDNP signal was not detected in the spectrum. These results are consistent with the solution structure obtained from the structure calculations; for example, all  three histidine residues are located on the surface of the LC3-I structure, and their side chains apparently have sufficient access to solvent (Fig. 1B). In contrast to the histidine residues, all four tyrosine residues are packed in the interior or are partially exposed to solvent according to the solution structure of LC3-I (Fig. 1B). Only the side chain of Tyr-38 is likely accessible to dye because this residue is included in the loop region connecting strands S1 to S2 and is located proximal to the disordered C-terminal region.
Role of the C-terminal Subdomain of LC3-I-Previous studies have shown that Atg8 is located in the precursor and mature structures of autophagosomes and that the modified form of Atg8 binds tightly to the autophagosome surface (11,13,19,24). Although the unmodified form does not show explicit partitioning into the detergent phase (13,19), LC3-I is abundant not only in the cytosol but also in the microsomal fraction (48). These reports imply that the LC3-I structure has a region that provides an interface with autophagosomes. To confirm this hypothesis, we performed a NMR titration experiment using DPC as a membrane mimic, and we attempted to identify the region that makes contact with DPC.
A number of HN cross-peaks in the HSQC spectrum of LC3-I shifted significantly when the LC3-I solution was titrated with DPC at concentrations ranging from 0.01 to 30 mM (Fig. 3A). All of these cross-peaks began to change at a concentration of ϳ0.1 mM and were perturbed in a dose-dependent manner up to 3.0 mM DPC (Fig. 3, A and B). The residues considerably affected by the addition of DPC (⌬␦ Ͼ 1.5 ppm) were Gly-40, Leu-47, Asp-48, Lys-51, Phe-52, Val-58, Asn-59, Leu-63, Arg-68, Leu-82, and Gly-85, the majority of which were located on the protein surface. These residues also appeared to be proximal to aromatic rings; for example, residues Leu-47, Asp-48, Lys-51, and Phe-52 were very close to the side chain of Phe-52 in the three-dimensional structure of LC3-I (Fig. 3C). Because the CD spectra of LC3-I with and without 5 mM DPC were almost identical (data not shown), it appears that the secondary structure of LC3-I is not affected by the addition of DPC. Therefore, these results suggest that the addition of DPC to the LC3-I solution leads to local structural alternations such as change in the orientation of aromatic rings. Interestingly, cross-peak perturbations were largely detected for the residues around strands S2 and S3 and helix H3, and it is even more noteworthy that the residues showing a chemical shift perturbation were clustered in the C-terminal subdomain of LC3-I (Fig. 3C). On the other hand, in residues 1 to 29, which constitute the Nterminal subdomain, the maximum change in ⌬␦ was 0.385 ppm for Leu-22, and most of the cross-peaks in the HSQC spectrum were scarcely perturbed by the addition of DPC (Fig.  3, B and C). Like the residues in the N-terminal subdomain, those located in the loop connecting helix H4 to strand S4 (H4/S4 loop) were also unaffected by the addition of DPC (Fig.  3C). For example, the chemical shift perturbation of residue Asp-104 was relatively small (the maximal change was 0.347 ppm; Fig. 3B). These findings indicate that the addition of DPC does not affect the conformation of the N-terminal subdomain and H4/S4 loop. In the H4/S4 loop, the side chains make van der Waals contacts with those of helix H2 and contribute to the formation of a hydrophobic core between the N-and C-terminal subdomains of LC3-I (Fig. 1B). In this core, the upfield-shifted resonances were observed for Ile-34 as described above. This upfield shift was almost completely retained even in the presence of DPC, indicating that the hydrophobic core between the N-and C-terminal subdomains is not affected by the addition of DPC. Whereas, the C-terminal subdomain also forms a distinct hydrophobic core in which some protons show an upfield shift (e.g. the 1 H ␦1 resonances of Ile-35 and Ile-67 are included). In strong contrast with the case of the core between the N-and C-terminal subdomains, these resonances significantly moved downfield upon addition of DPC; for example, the chemical shift value of Ile-67 1 H ␦1 changed from 0.183 to 0.496 ppm. These downfield perturbations are indicative of environmental changes in the hydrophobic core involved in the C-terminal subdomain. Thus, the NMR experiments with DPC showed that only the C-terminal subdomain of LC3-I is sensitive to DPC.  (49). Another mammalian homologue of Atg8, GABARAP, also interacts with purified tubulin and tubulin-polymerized microtubules. This interaction requires the region of GABARAP that corresponds to the Nterminal subdomain of LC3-I (25,31). Although neither MAP-LC3 nor GABARAP has the microtubule-binding domain found in MAP1A and MAP1B, charged residues are clustered in their N-terminal subdomains and form extensive polar surfaces. Therefore, it is expected that LC3-I binds to tubulin and/or tubulin-polymerized microtubules via the N-terminal subdomain. We therefore tested the binding of LC3-I to tubulin and microtubule proteins using wild-type LC3-I and a truncated mutant (LC3-I⌬N) lacking the N-terminal subdomain (residues 1 to 29).

Role of the N-terminal Subdomain of LC3-I-Previous studies have shown that MAP-LC3 binds to microtubule proteins via electrostatic interactions
Prior to conducting the binding analyses, we performed preliminary NMR experiments to confirm the structural integrity of LC3-I⌬N. The 1 H ␣ resonances of Val-112 and Tyr-113 showed extreme downfield shifts (6.033 and 6.162 ppm, respectively) in the wild-type LC3-I structure. Similar features were still observed for LC3-I⌬N (5.889 and 6.121 ppm, respectively). Both of these residues are included in strand S4 of the central ␤-sheet, and the conservation of the chemical shifts indicated that the LC3-I structure was maintained. In addition, LC3-I⌬N gave the NOE connectivities characteristic of a ␤-sheet; for example, the NOE signal was observed between the two 1 H ␣ protons of Tyr-113 and Leu-81 that belong to strands S4 and S3, respectively. These results strongly suggest that LC3-I⌬N retains the original structure, at least within the central ␤-sheet, found in the C-terminal subdomain of wild-type LC3-I.
Injection of wild-type LC3-I onto a sensor chip with immobilized tubulin caused a dose-dependent increase in the surface plasmon resonance analysis response (Fig. 4A). In addition, the bound LC3-I protein was easily released in the presence of 400 mM NaCl (Fig. 4A, inset). These observations indicate that LC3-I binds to the purified tubulin and that the interaction between LC3-I and tubulin is largely mediated by electrostatic interactions. In contrast to the wild-type LC3-I, LC3-I⌬N showed little tubulin binding activity (Fig. 4A). Taken together, these results suggest that the N-terminal subdomain of LC3-I is essential for the interaction between LC3-I and tubulin.
Co-sedimentation assays were carried out using the wildtype and the truncated LC3-I constructs to verify the importance of the N-terminal subdomain in the binding of LC3-I to microtubule proteins. As shown in Fig. 4B, LC3-I co-sedimented with microtubules that were assembled from purified tubulin in the presence of paclitaxel (Fig. 4B, lane 1). This interaction between LC3-I and microtubules disappeared upon addition of 400 mM NaCl (Fig. 4B, lane 3). On SDS-PAGE, LC3-I⌬N migrated more rapidly than the wild-type LC3-I by ϳ5 kDa because of the deletion of the N-terminal subdomain. The co-sedimentation assay using the truncated form of LC3-I gave only a faint band at the position corresponding to LC3-I⌬N (Fig. 4B, lane 2). Thus, the co-sedimentation assay showed that the N-terminal subdomain of LC3-I plays an essential role in microtubule binding and that the interaction between LC3-I and microtubules is governed by electrostatic effects. DISCUSSION Before the modification by PE, MAP-LC3 (LC3-I) adopts a conformation similar to that of the other Atg8 homologues, GABARAP and GATE-16, and its structure can be divided into N-terminal (residues 1 to 29) and C-terminal (residues 30 to 120) subdomains. The N-terminal subdomain of LC3-I is composed of short (H1) and long (H2) ␣-helices. In the case of GABARAP, the region corresponding to helix H1 of LC3-I adopts distinct "closed" and "open" conformations (31). Whereas the N-terminal 10 residues of GABARAP form a 3 10 helix and cover up the surface of the C-terminal subdomain in the closed conformation, this region is projected away from the C-terminal subdomain in the open conformation. The photo-CIDNP analysis of LC3-I revealed that the resonance derived from Tyr-110 shows no CIDNP effects, which clearly indicates that the side chain of Tyr-110 is prevented from interacting with the photoexcited dye. This result, taken together with the LC3-I structure, supports the conclusion that this interference is because of the presence of the helix H1. Therefore, LC3-I adopts a single compact conformation corresponding to the closed conformation of GABARAP, and the helix H1 of LC3-I covers up the side chains, including Tyr-110.
Titration with DPC caused a dose-dependent chemical shift perturbation for some residues of LC3-I, which indicates that LC3-I associates with DPC. Considering that the critical micelle concentration of DPC is ϳ1 mM (50), the change of LC3-I caused by the addition of DPC is probably accompanied by the formation of DPC micelles. Previous studies have demonstrated that the chemical shifts derived from membrane-binding proteins are significantly perturbed by titration with DPC (51). Therefore, the results of DPC titration suggest that LC3-I recognizes the DPC micelles but not a single molecule of DPC. Although these interpretations might conflict with the reports that LC3-I does not show explicit partitioning into the detergent phase (13,19), our data can be explained by assuming that LC3-I interacts only weakly with the micelles. The tight binding of MAP-LC3 to autophagosomes, on the other hand, requires that PE attached covalently to the C terminus (i.e. LC3-II form).
The LC3-I structure can be divided into two regions with distinct sensitivities to DPC: the C-terminal subdomain, especially strands S2 and S3 and helix H3, is sensitive to DPC, whereas the N-terminal subdomain and the H4/S4 loop regions are insensitive to DPC. This difference might be explained by a lack of uniformity of the surface charge distribution of LC3-I. The N-terminal subdomain and the H4/S4 loop have extensively charged surfaces because they include 16 residues that are charged at neutral pH. In contrast, some apolar regions were found on the surface of the C-terminal subdomain. For example, the side chains of nonpolar residues Ile-64, Ala-75, Ala-78, Phe-80, Met-88, Val-89, Val-91, Phe-119, and Gly-120 were exposed to solvent, and they formed an extensive apolar surface around strand S3 of LC3-I. These surfaces would give LC3-I the ability to associate with DPC acyl chains via hydrophobic interactions. In addition, some polar surfaces located in the C-terminal subdomain might be involved in the association of LC3-I with the charged head group of DPC. Although the detailed mechanism of interaction between LC3-I and the micelles remains unclear, it appears LC3-I associates with DPC micelles at least via its C-terminal subdomain. This implies that the modified form of MAP-LC3 (LC3-II) also uses the C-terminal subdomain as an interface for interacting with autophagosomes.
The binding assays clearly demonstrated that the N-terminal subdomain of LC3-I is indispensable for binding both purified tubulin and microtubules. In general, microtubules are known to be formed by ␣and ␤-tubulin pairs. Both of those two isotypes of tubulin have a number of acidic residues at the C-terminal region that contribute to the binding between MAPs and microtubules (52)(53)(54)(55). Structural analyses have indicated that ␣and ␤-tubulin form basically identical structures, and their acidic C-terminal regions are located on the outside surface of microtubules (56,57). Such studies have provided evidence that the basic regions of MAPs and their light chains associate with the C-terminal segment of tubulin via electrostatic interactions. The N-terminal subdomain of LC3-I also possesses some basic residues that are exposed to solvent, such as Lys-5, Lys-8, Arg-10, Arg-11, Arg-21, and Arg-24. These residues are therefore likely to be involved in the binding to tubulin and microtubules. In addition, Coyle et al. (31) showed that residues 10 to 22, which are included in helix H2 of GABARAP, form the minimal tubulin-binding region. Like the N-terminal subdomain of LC3-I, this region of GABARAP includes some basic amino acids. Moreover, these basic residues in GABARAP are located on the surface of one side of the helix H2 and play an important role in the binding to tubulin (31). Although LC3-I has some basic amino acids in its N-terminal subdomain, only Arg-21 and Arg-24 are located on the surface of the region that corresponds to the minimal tubulin-binding site of GABARAP. In the LC3-I structure, the region including helix H1 has a more extensive, positively charged surface (because of the presence of Lys-5, Lys-8, Arg-10, and Arg-11) than helix H2 of LC3-I. This suggests that the basic residues located around helix H1 contribute to the interactions of LC3-I with tubulin and microtubules.
Taken together, the N-terminal subdomain shows an extensively charged surface and contributes to the binding of MAP-LC3 to tubulin and microtubules. Moreover, our titration experiments suggested that the C-terminal subdomain makes contacts with the membrane surface of autophagosomes, whereas the N-terminal subdomain is oriented toward solvent. These features of MAP-LC3 may allow it to act as an adaptor protein between microtubules and autophagosomes, an idea consistent with previous reports on the intracellular localization and the binding activity of MAP-LC3 (11,13,19,21,24). Conversely, it has been reported that autophagy in yeast is not dependent on microtubules because nocodazole, a microtubuledepolarizing agent, was shown to have no effect (11). Furthermore, Lang et al. (58) proposed that Atg8 attaches indirectly with microtubules through another Atg protein, Atg4. These conflicting findings may be because of the difference between yeast and mammalian cells. It is likely that microtubules are not absolutely essential for autophagy, although they may assist in efficient formation and/or intracellular trafficking of autophagosomes. GABARAP, another mammalian homologue of yeast Atg8, acts as an adaptor protein between the scaffold protein gephyrin and the GABA A receptor and participates in intracellular trafficking and/or receptor anchoring to the appropriate cell surface (25, 59 -61). The N-terminal subdomain of GABARAP is essential for its binding to microtubules, and the C-terminal subdomain provides the interface necessary for the association with GABA A receptor (25,30,31,59,60). Here, we showed the high degree of structural similarity between LC3-I and GABARAP, which strongly indicates that MAP-LC3 and GABARAP have similar functions. Interestingly, MAP-LC3 has unique binding partners, including not only protein components but also autophagosome membrane (11,13,19,24). Our titration analysis with DPC suggested that the Cterminal subdomain of LC3-I is involved in the association with the micelles and mediates the recognition of autophagosomes by MAP-LC3.
Although autophagy is known to require many factors including Atg8, the structural mechanism remains to be elucidated. MAP-LC3 levels correlate directly with the autophagic activity of mammalian cells (21), implying that MAP-LC3 plays an essential role in the process and regulation of autophagy. The present analyses identified the functional subdomains in the LC3-I structure and suggested that MAP-LC3 acts as an adaptor protein between microtubules and autophagosomes. Moreover, MAP-LC3 interacts with other proteins such as human homologues of Atg7 and Atg3 (18,22). This feature is also essential for the association of a set of Atg members with autophagosomes. However, it remains unclear how autophagosomes are generated by Atg family proteins, especially from the point of view of the molecular structure. Although further analyses will be needed to elucidate the structural mechanism of autophagosome formation and intracellular trafficking, the solution structure and the identification of the functional subdomains of LC3-I will provide useful information for a molec-ular understanding of protein-membrane interactions and membrane-mediated transport.