Development of new tools to study lipidated mammalian ATG8


 Mammals conserve multiple mammalian ATG8 proteins (mATG8s) consisting of γ-aminobutyric acid receptor-associated protein (GABARAP) and microtubule-associated protein 1 light-chain 3 (LC3) subfamilies that tightly bind to the autophagic membranes in a lipidated form. They are crucial in selective autophagy and recruit proteins bearing LC3-interacting region (LIR) motifs. However, because limited research tools are available, information about the specific roles of each lipidated mATG8 in selective autophagy is scarce. Here, we identified LIR motifs specific to the lipidated form of each mATG8 and characterized the residues critical for their selective interaction using cell-based assays and structural analyses. Then, we used these selective LIR motifs to develop probes and irreversible deconjugases that targeted selective lipidated mATG8s in the autophagic membrane, revealing that lipidated GABARAP subfamily proteins regulate aggrephagy of amyotrophic lateral sclerosis-linked protein aggregates. Our tools will be useful in elucidating the functional significance of each mATG8 protein in autophagy research.


INTRODUCTION
Autophagy is an evolutionarily conserved cellular degradation pathway that selectively or nonselectively eliminates unwanted materials, such as damaged organelles and harmful cytosolic aggregates, to protect the cell's ability to regulate homeostasis, adapt to various stressors, differentiate during development, and prevent genomic damage 1 . Recent studies on the role of autophagy in secretion and exocytosis have expanded our understanding of its biological significance 2,3 . Furthermore, autophagy dysfunction has been linked to several human diseases, including cancer, neurodegenerative diseases, infectious diseases, liver diseases, and cardiovascular disorders 4,5 .
The autophagy process is tightly regulated by many autophagy-related (ATG) proteins.
To date, more than 40 ATG genes have been identified in yeast and higher eukaryotes, constituting a diverse family of genes whose products not only precisely control autophagy but also play roles in membrane trafficking and signaling pathways 1 . ATG8 is a small ubiquitinlike protein in yeast that is covalently conjugated to phosphatidylethanolamine (PE) in autophagosomes following proteolytic cleavage of its C-terminus by the cysteine protease ATG4, which is also involved in the delipidation of ATG8-PE to release ATG8 from autophagosomes. This cleavage is necessary to elongate and close the phagophore membrane.
Although there is only one ATG8 protein in yeast, mammals have two subgroups: the family of microtubule-associated protein light-chain 3 (LC3) proteins (LC3A, LC3B, and LC3C) and the family of γ-aminobutyric acid receptor-associated proteins (GABARAP, GABARAP-L1, and GABARAP-L2) 6 . These proteins undergo reversible lipidation by PE conjugation to their C-terminal regions in the autophagosome. Although mammalian ATG8 protein (mATG8) conjugation is crucial for conventional autophagy processes, including autophagosome biogenesis and maturation, accumulating evidence also suggests its involvement in nonconventional autophagy processes, such as secretory autophagy, LC3-associated phagocytosis, entosis, micropinocytosis, and LC3-associated endocytosis 7 . However, the biological relevance of mATG8 diversity in conventional and nonconventional autophagy or even in autophagy-independent pathways is largely unknown. mATG8 proteins recruit autophagic machinery that contains LC3-interacting region (LIR) motifs to autophagosomes 8,9 . They can also sequester selective cargos into autophagosomes via LIR motif-containing receptors or adaptors during selective autophagy 10 .
Therefore, the specific roles of LC3 and GABARAP subfamily proteins are regulated by many LC3-and GABARAP-interacting proteins with LIR motifs. Many mATG8-interacting proteins contain canonical LIR motifs that have a hydrophobic LIR motif with a core consensus sequence of (W/F/Y)-X-X-(L/I/V) that binds to the W-site and L-site conserved in mATG8 proteins using the amino acids W/F/Y and L/I/V, respectively, in the consensus sequence 8,11,12,13,14 . The GABARAP-selective motif was recently proposed to have a core consensus sequence of (W/F)-(I/V)-X-V, similar to that of an LIR motif 15 . However, some mATG8-interacting proteins contain non-canonical LIR motifs that do not meet the sequence requirements for canonical LIR motifs and present different structural determinants that are involved in mATG8 interactions 16,17 . Many LIR motifs that interact with mATG8 proteins have been extensively investigated in recent studies 9,18,19 . However, the selective binding mechanism remains largely unknown. Several studies have used peptide-based arrays, glutathione S transferase (GST) pulldown assays, or competitive time-resolved fluorescence resonance energy transfer to examine the binding properties of LIR motifs for mATG8 9,18,20,21,22 . Although these approaches are useful, the assays are purely in vitro and may reflect the binding property of LIR motifs of only non-lipidated mATG8 proteins. Thus, many in vitro assay results may not reflect the physiological binding properties of lipidated mATG8 proteins on the autophagic membrane in cells. Therefore, identification of the functional LIR motifs and characterization of the determinants that result in selective binding to lipidated mATG8 proteins on the autophagic membrane in cells are crucial.
In this study, we used novel cellular assays to determine the binding properties of LIR motifs to mATG8 proteins in cells. We utilized structural analyses to identify LC3C-, GABARAP-L2-, and GABARAP/GABARAP-L1-selective LIR motifs and characterize the biochemical properties responsible for their selective binding. Furthermore, we generated a new system that selectively monitors or delipidates lipidated mATG8s in the autophagic membrane by incorporating identified selective LIR motifs into RavZ, an irreversible deconjugase for mATG8. Finally, the use of these selective probes or deconjugases for mATG8-PE revealed that the GABARAP subfamily proteins regulate cellular degradation of amyotrophic lateral sclerosis (ALS)-linked TDP25 (the 25-kDa C-terminal fragment of TDP-43) protein aggregates during aggrephagy. These data demonstrate that our newly developed tools are widely applicable to elucidating the functional significance of the lipidated form of each mATG8 protein in diverse autophagic and non-autophagic processes.

Characterization of Binding Properties of LIR Motifs Toward Lipidated mATG8 in Hexa mATG8-knockout HeLa (HKO) Cells.
To examine the binding properties of LIR motifs for each mATG8 and to identify selective mATG8-binding LIR motifs in cells, we first used an LIR motif tagged with monomeric red fluorescent protein (mRFP) fused to a 3xNLS sequence (LIR-mRFP-3xNLS) for a nuclear localization signal or sequence (NLS) assay in HKO cells, in which endogenous mATG8s are not expressed (see detailed information in the STAR Methods section) ( Figure S1A; Figure  1A). We first examined the binding properties of an LIR motif from p62 (LIR(p62)), the first LIR motif for mATG8 proteins that was identified in the autophagic membrane 23 . Each green fluorescent protein (GFP)-mATG8(GA) protein (in which the C-terminal glycine residue was replaced with alanine to impair PE conjugation [lipidation] and affect their cellular localization) was mainly localized to the nucleus in cells expressing LIR(p62)-mRFP-3xNLS but not the LIR motif mutant LIR(p62)m-mRFP-3xNLS (Figures 1B). The relative GFP fluorescence intensity ratio between the nucleus and cytosol (N/C ratio) in LIR(p62)-mRFP-3xNLSexpressing cells was significantly higher than that in LIR(p62)m-mRFP-3xNLS-expressing cells for all mATG8s (Figures 1C), indicating that this LIR bound to all types of non-lipidated mATG8 in cells in an LIR-dependent manner. Consistent with the NLS assay, LIR(p62)-GFP but not LIR(p62)m-GFP showed binding with all mATG8 proteins in the GST pull-down and co-immunoprecipitation (co-IP) experiments using GST-mATG8 or 3xFLAG-mATG8 proteins, respectively ( Figures 1D and 1E), confirming that LIR(p62) bound to all mATG8 proteins.
To validate the binding property of the p62 LIR motif to lipidated mATG8 proteins, we expressed LIR(p62)-GFP and mRFP-mATG8 in HKO cells and quantified the relative GFP fluorescence intensity ratio between the autophagosome and cytosol (A/C ratio) of LIR(p62)-GFP ( Figure 1F). Unexpectedly, LIR(p62)-GFP mostly localized to mRFP-LC3 subfamilypositive autophagic membranes compared with mRFP-GABARAP subfamily-positive autophagic membranes in HKO cells, which was greatly reduced for LIR(p62)m-GFP ( Figures   1G and 1H), indicating that LIR(p62)-GFP was preferentially localized to LC3 subfamilypositive autophagic membranes. These results suggested that LIR(p62) could interact with all types of non-lipidated mATG8s in the soluble condition but was targeted to the autophagic membrane, probably via preferential binding to lipidated LC3 subfamily proteins.
Our results suggested that LIR motifs might have different binding properties to nonlipidated mAtg8 in solution and lipidated mATG8 on the autophagic membrane. Therefore, we further investigated and characterized the binding properties of other LIR motifs using our two alternative assay systems: GST pull-down and NLS assays. Both assays showed that an LIR motif from TBC1D25 (LIR(TBC1D25)) or FAIM2 (LIR(FAIM2)) bound to all or none of the mATG8 proteins, respectively (Figures S1B-S1E). In contrast, both of them showed preferential localization to mRFP-LC3 subfamily-positive autophagic membranes in HKO cells (Figures S1F and S1G). Thus, our results showed that some LIR motifs had differential binding properties for non-lipidated and lipidated mATG8 proteins in cells.

Identification of Selective Lipidated mATG8-binding LIR Motifs
Since our goal was to identify selective LIR motifs for lipidated mATG8 proteins, we next examined the binding properties of various GFP-fused known LIR motifs with mRFP-mATG8 proteins in HKO cells (Table 1; Figure 2). To this end, we expressed each LIR(X)-GFP (X: the protein name from which the LIR motif originated) together with each mRFP-mATG8 and quantified the A/C ratio of each mATG8 in HKO cells upon starvation in the presence of bafilomycin A1 (BafA1).
Next, we identified the LIR motif that preferentially localized to GABARAP-L1positive autophagic membranes ( Figures 2C and 2D). An LIR motif from Nix/BNIP-3L (LIR(Nix)) showed no significant binding to any mATG8 (Table 1). However, a phosphomimetic mutant of Nix/BNIP-3L (LIR(Nix-p)), in which two serine resides located Nterminal to the core LIR motif were substituted by phosphomimetic E residues, showed selective localization to the GABARAP-L1-positive autophagic membrane 25 ( Figure 2C). On the other hand, in the GST pull-down assay, LIR(Nix-p) interacted with all mATG8 proteins ( Figure 2D). These results demonstrated that LIR(Nix-p) could bind to all non-lipidated mATG8 proteins but was selective for lipidated GABARAP-L1-positive autophagic membranes in cells.
Additionally, we identified the LIR motif that preferentially localized to GABARAPand GABARAP-L1-positive autophagic membranes ( Figures 2E and 2F). The LIR motif from the C-terminus of ATG4B (LIR(4B)) showed binding to all lipidated mATG8 proteins except for LC3C (Table 1). A previous two-dimensional peptide array scan analysis showed that the replacement of F with S/T within the core region of LIR(4B) severely impaired binding to LC3B but not to GABARAP 26 . Consistent with this report, a mutation of F to T within the core region of LIR(4B) (referred to as LIR(4B(T)), which makes this motif atypical, almost diminished binding to the LC3 subfamily but showed significant binding to nonlipidated/lipidated GABARAP and GABARAP-L1 ( Figures 2E and 2F). Thus, LIR(4B (T)) showed selective binding to GABARAP and GABARAP-L1.

Structural Basis for Selective Binding of mATG8-binding LIR Motifs
We attempted to solve the crystal structures of atypical LIRs bound to mATG8 proteins to reveal the specific mechanism of LIR interaction with mATG8 proteins at the atomic level and succeeded in determining the structure of the LIR(Sp)-GABARAP-L2 fusion protein at a resolution of 1.86 Å ( Figure 3A; Table S1). The structure of the LIR(Sp)-GABARAP-L2 complex was similar to that of the LIR(Sp)-SpAtg8 complex ( Figure S3A). SpHfl1 LIR consisted of an α-helix, from D391 to M404, and an N-terminal tail. The helix forms extensive hydrophobic interactions with V51, P52, I55, W62, and I63 of GABARAP-L2 using M394, L397, Y398, A401, and M404. Among these, Y398 forms the most critical interaction by deeply inserting into the L-site pocket of GABARAP-L2. In addition to the hydrophobic interactions, three acidic residues, namely, D391, E393, and E395, formed electrostatic interactions with K46, R67, and R28 in GABARAP-L2, respectively ( Figure 3B) 24 . Although most interactions were similar to those in the LIR(Sp)-SpAtg8 structure, F388 of SpHfl1 was not inserted into the W-site of GABARAP-L2. Considering that alanine substitution of F388 induced a limited decrease in the binding affinity to SpAtg8, this observation implied that the W-site binding was unimportant for SpHfl1 binding to ATG8-family proteins. We also performed a sequence alignment of mATG8 proteins ( Figure S3B) 24 . Among the GABARAP-L2 residues involved in the interaction with LIR(Sp), W62 was the sole residue that was not conserved in the other mATG8 proteins (F, K, or S was also observed at this position). To assess the importance of W62 for this interaction, we performed ITC experiments using three mATG8 mutants (GABARAP-L2 W62A, GABARAP F62W, and LC3B K65W) ( Figure 3E).
The W62A mutation in GABARAP-L2 reduced the binding affinity with LIR(Sp), but the F62W mutation in GABARAP had the opposite effect. The K65W mutation in LC3B marginally increased this interaction. Coupled with the fact that SpAtg8, which binds strongly with SpHfl1, has a Y at position 62, these data suggested that either W or Y at position 62 was necessary, but not sufficient, for the strong binding of ATG8 family proteins with SpHfl1 LIR.
The crystal structure of the LIR(TP)-mATG8 complex has not been previously reported. We failed to crystallize the LIR(TP) mutants bound to mATG8 proteins but succeeded in crystallizing and determining the structure of the wild-type LIR(TP)-GABARAP fusion protein at a resolution of 1.75 Å ( Figure 3C; Table S1). The conformation of LIR(TP) was unique compared with that of canonical LIRs, whereas W35 and I38 bound to the W-site and L-site, respectively, in a canonical manner, the region N-terminal to the core LIR sequence (residues 28-33) formed an intramolecular β-sheet with the core LIR sequence ( Figure 3C, left). An intramolecular but distinct β-sheet was also observed for the RavZ LIR, with the region C-terminal to the core LIR sequence forming a β-sheet with the core LIR sequence 27 .
Although the LIR(TP) peptide used for crystallization possesses six acidic residues, only one residue (E29) formed an electrostatic interaction with GABARAP (R67), suggesting that the binding affinity was largely dependent on the core LIR motif. To address the specificity of LIR(TP(T)) to LC3C, we prepared a structural model of the LIR(TP(T))-LC3C complex by superimposing the wild-type LIR(TP)-GABARAP structure onto the LC3C-NDP52 structure (Protein Data Base ID [PDBID]: 3VVW) followed by manual model adjustment ( Figure 3D) 16 . NDP52 LIR possessed a non-canonical core sequence (I 133 -L-V-V 136 ) and showed specific interaction with LC3C, with V136 binding to the L-site, whereas I133 did not bind to the Wsite. The lack of canonical interaction must be compensated by additional interactions to maintain the high binding affinity, including hydrophobic interactions between V135 and LC3C F33 and a hydrogen bond between N129 and LC3C K32. In the case of LIR(TP(T)), I37 and D33 could form hydrophobic and electrostatic interactions with LC3C F33 and K32, respectively. Additionally, E29 and E31 of LIR(TP(T)) formed electrostatic interactions with R76 and K55 of LC3C, respectively. Among these residues, K55 and R76 were strictly conserved in all mATG8 proteins, and K32 was conserved within the GABARAP subfamily, whereas F33 was unique to LC3C. These observations suggested that F33 of LC3C was responsible for its observed specificity. Consistent with this, the mutation of K32/F33 to Q/H (corresponding to LC3A or LC3B) but not Y (corresponding to the GABARAP subfamily) reduced the binding affinity of LC3C to LIR(TP(T)) ( Figure S1H). We noticed that the α2 helix of the LC3 family, whose C-terminus contained K32 and F33, was located closer to the LIRbinding pocket than that of GABARAP subfamily proteins ( Figure S3C, left), a feature that seemed to enable K32 and F33 of LC3C to interact with D33 and I37 of LIR(TP(T)) ( Figure   3D). The distinct positioning of the α2 helix could be attributed to the type of amino acid at position 18 (using GABARAP numbering). The LC3 family possessed a V at this position, which had a larger side-chain than G (GABARAP/GABARAP-L1) and S (GABARAP-L2), resulting in a steric crush of the α2 helix with the ubiquitin fold, thereby positioning the α2 helix toward the LIR-binding site ( Figure S3C, right). Consistent with this, the mutation of V26 to G (corresponding to GABARAP and GABARAP-L1) reduced the binding of lipidated LC3C to LIR(TP(T)) (Figures S1H). Therefore, the combination of K32, F33, and the properly positioned α2 helix was likely necessary for the specific binding of LC3C to LIR(TP(T)).

Development of Probes Selectively Monitoring Lipidated mATG8 Proteins on Autophagosomes
Next, we monitored mATG8-positive autophagic membranes to study the function of each mATG8 protein in autophagy or selective autophagy using each identified selective lipidated mATG8-binding LIR motif. To this end, we replaced the LIR1/2 and LIR3 motifs within RavZ(ΔCA)-GFP (termed gProbe, g: GFP) with selective lipidated mATG8-binding LIR motifs (gProbe-X, X: the protein name, from which the LIR motif originated), as described previously 28 . Each probe, which contained two LIR motifs and an MT domain, was coexpressed with each of the mRFP-mATG8 proteins in HKO cells ( Figure 4A) and quantified the A/C ratio upon starvation in the presence of chloroquine (Table S2 and Figure 4B).
Among the selective LIR motifs that we characterized, LIR(Sp) was highly selective for GABARAP-L2 (Figure 2A and 2B). Therefore, we first tested whether gProbe-Sp was selectively localized to GABARAP-L2-positive autophagic membranes. Indeed, gProbe-Sp was mostly localized to GABARAP-L2-positive autophagic membranes; however, it was less but significantly localized to other types of mATG8-positive autophagic membranes, indicating that gProbe-Sp could detect all types of mATG8-positive autophagic membranes at differential levels ( Figure 4B). L397 in LIR(Sp) hydrophobically interacted with W62 in GABARAP-L2, which corresponds to F62 in GABARAP/GABARAP-L1 ( Figure 3). Therefore, we considered L397 a good candidate for a mutation to generate a more selective GABARAP-L2-binding LIR motif. We replaced L397 with I (LIR(Sp(I)) to generate gProbe-Sp(I), which was selectively localized to GABARAP-L2-positive autophagic membranes ( Figure 4B), suggesting that gProbe-Sp(I) is a highly selective probe for GABARAP-L2positive autophagic membranes in cells.
Finally, we confirmed LC3 subfamily-or GABARAP subfamily-positive autophagic membrane targeting using previously developed gProbe-Fy or gProbe-St, respectively 28 , in HKO cells (Table S2) Figures 4C-4G). Notably, some autophagosomes were only detected by LC3 subfamily-selective rProbe-Fy ( Figure 4G). Taken together, these data suggested that our new selective LIR-based probes were useful to identify autophagosomes containing distinct mATG8 subfamilies at the cellular and ultrastructural levels.

Development of Enzymes Selectively Delipidating mATG8-PE in Autophagic Membranes
To date, due to limitations in tools that can selectively inhibit or deplete each lipidated mATG8 protein, information concerning the specific roles of the lipidated and non-lipidated forms of each mATG8 protein in autophagy and in autophagy-independent pathways is scarce.
Compared with mammalian ATG4B, which hydrolyzes the amide bond linking glycine and PE, RavZ hydrolyzes the amide bond between the C-terminal glycine residue and an adjacent aromatic residue, resulting in resistance to conjugation by the host machinery 29,30 . Therefore, to generate deconjugase to selectively remove PE from mATG8-PE in our study, we replaced and GABARAP subfamily proteins. On the other hand, the NDP52 level was unaffected by Deconjugase-Fy expression ( Figure 5F), suggesting that GABARAP subfamily proteins but not LC3A/B regulated the autophagic degradation of NDP52.

Lipidated GABARAP Subfamily Proteins Regulate TDP25-mediated Aggrephagy
We attempted to elucidate the specific roles of mATG8 proteins in aggrephagy, a type of selective autophagy 10 , using our selective mATG8 LIR-based probes and deconjugases.
Although several aggregate-prone proteins play roles in many different human diseases, including neurodegenerative diseases, and may be cleared via aggrephagy, little is known about the differential roles of lipidated LC3 or GABARAP subfamily proteins in aggrephagy.
Thus, we investigated which mATG8 proteins were involved in the autophagic degradation of TDP25 aggregates. TDP25 is a pathogenic aggregate-prone 25 kDa C-terminal protein of TDP-43 that has been identified in protein inclusions in several neurodegenerative diseases, including frontotemporal dementia and ALS 31,32 . First, we transfected Myc-TDP25 into cultured cortical neurons to verify whether Myc-TDP25 aggregates were degraded by autophagy. TDP25-positive aggregates were observed 24-48 h after transfection. When autophagy was activated with trehalose, an mTOR-independent autophagy inducer, the size of the TDP25-positive aggregates was significantly reduced, but their number was unaffected ( Figure 6). However, when autophagy was inhibited by BafA1 in neurons expressing Myc-TDP25 upon autophagy induction, the size of TDP25 aggregates increased, suggesting that the reduced size of TDP25 aggregates might be due to autophagic degradation of TDP25.

DISCUSSION
Most binding experiments with ATG8 proteins have been performed with soluble, non-lipidated/free mATG8 proteins. However, the cellular functions of mATG8 proteins frequently occur on the autophagic membranes in cells when mATG8 is in the lipidated form. To date, there is limited knowledge regarding the differences in binding between the non-lipidated and the lipidated forms. In this study, we used our novel cellular assays to demonstrate that lipidated and non-lipidated mATG8 proteins could possess differential binding properties for the same LIR motifs. Our cellular assay showed that LIR(p62) or LIR(FAIM2) preferentially bind to the lipidated forms of LC3 subfamily proteins rather than GABARAP subfamily proteins on autophagic membranes, although they bind to all or none of the mATG8s in the non-lipidated form, respectively (Figures 1 and S1B-S1G). It has been consistently reported that p62 binds to both LC3B and GABARAP-L2 in the cytosol but only to LC3B in autophagic membranes, whereas FAIM2 only binds to lipidated LC3B (LC3B-II) via the LIR motif using co-IP experiments in cells 33,34 . Why then do some LIR motifs show different binding properties to mATG8 depending on its localization? It is possible that the α1 helix of mATG8s is primarily involved in these binding differences. The α1 helix of yeast Atg8 undergoes a conformational change upon lipidation 35 . Interestingly, the substitution of the N-terminal α1 helix between LC3B and GABARAP-L2 reversed the recruitment efficacy of p62 to the autophagic membrane 33  Although the specific roles of each mATG8 protein in selective autophagy are mainly unknown, it is known that ATG8 is recruited together with cargo into autophagosomal membranes via interaction with autophagy receptors to facilitate cargo degradation.
Interestingly, most autophagy receptors possess an LIR motif that allows their direct binding to LC3, whereas most autophagy adaptor proteins have GABARAP-specific LIRs. Therefore, it has been proposed that LC3 subfamily proteins are essential for cargo recruitment upon selective autophagy 7  Additionally, based on our cellular assay, many LIR motifs selectively bind to GABARAP subfamily proteins compared with LC3A/B (Table S1). This might be because many different LIR-containing autophagy receptors or other autophagy machinery proteins including fusion components selectively bind to GABARAP subfamily proteins to regulate selective targeting or autophagic degradation. However, further detailed cellular and molecular approaches are necessary to elucidate the exact mechanism that regulates the selective interaction of each GABARAP subfamily protein and autophagy component.
In conclusion, our LC3-and GABARAP-selective LIR-based sensors and irreversible selective deconjugases for each mATG8-PE on autophagosomes allow elucidation of the cellular localization and selective functions in conventional and nonconventional autophagy associated with these lipidated mATG8 proteins.

DNA Constructs
The sequence encoding 3xNLS was generated by polymerase chain reaction (PCR) amplification of C1-pEGFP-NUC vectors and inserted into an N3-mRFP vector to generate N3-mRFP-3xNLS using the restriction enzyme set Xho1-Not1. All LIRs, including the mutant LIRs used in these experiments, were amplified by extension PCR without a template using primers and then inserted into N3-EGFP and N3-mRFP-3xNLS vectors using the restriction enzyme set HindIII-Kpn1. We used an extended LIR motif with 10 N-terminal amino acids and 11 C-terminal amino acids in addition to the core LIR motif sequence (W/F/Y)-X-X-(L/I/V) because N-terminal and C-terminal amino acids may also contribute to mATG8 protein binding 8,18 . Mutagenesis of the LC3 or GABARAP subfamily was amplified by PCR using

Quantitative Analysis of N/C Ratios
If an LIR motif interacts with a GFP-tagged mATG8 in the cytosol, the LIR-mRFP-3xNLS should sequester the cytosolic mATG8 in the nucleus, depending on the binding preference of the LIR-mRFP-3xNLS protein in HKO cells, in which endogenous mATG8s are not expressed ( Figure S1A). This enables the quantification of the relative GFP fluorescence intensity ratio between the nucleus and cytosol (N/C ratio) as well as the calculation of the N/C ratio in live cells ( Figure 1A). To test this, we used GFP-tagged mATG8 mutants (GFP-mATG8(GA)), in which the C-terminal glycine residue was replaced with alanine to impair PE conjugation (lipidation) and affect their cellular localization, as compared with a GFP-tagged wild-type mATG8 protein localized to autophagosomes. To express the quantitative ratio of the N/C fluorescence intensities, the average value of the nuclear and vesicular fluorescence intensities was measured by averaging at least five randomly selected points in the nucleus and the cytosol in a single cell using ImageJ software. In the same manner, the quantitative N/C ratio of at least five randomly selected cells was analyzed. The obtained values were normalized to the values calculated from mRFP-3xNLS-expressing cells. All statistical data were calculated and plotted using GraphPad Prism 6. We also confirmed that the expression of mRFP-3xNLS lacking an LIR motif did not affect the N/C distribution of GFP-mATG8(GA).

Quantitative Analysis of A/C Ratios
If an LIR motif interacts with a lipidated mRFP-mATG8 on the autophagic membrane, its localization from the cytosol to the mRFP-mATG8-positive autophagic membrane in cells will be observed ( Figure 1F). Therefore, if the ratio of GFP fluorescence intensity between the autophagic membrane and cytosol (A/C ratio) is compared, the relative binding affinity of an LIR motif for each expressed mRFP-mATG8 on the autophagic membrane can be determined.
To calculate the ratio of autophagosome/cytosol (A/C) fluorescence intensities in data obtained from fixed HKO cells after inducing and blocking autophagy flux by treatment with rapamycin and BafA1, the average value of the autophagosome or cytosol fluorescence intensity was obtained from at least five randomly selected points in autophagosomes or in the cytosol of a single HKO cell using ImageJ. In the same manner, the quantitative A/C ratio of at least 20 randomly selected cells per experiment was obtained from three independent experiments. All statistical data were calculated and plotted using GraphPad Prism 6.

GST Pull-down Assay
For the GST pull-down assay using HEK293T cell lysates, the cells were transfected with plasmid DNA encoding GFP constructs using calcium phosphate transfection. After

Correlative Light and Electron Microscopy
CLEM was performed as previously described 42  EMBed-812 embedding kit. The embedded samples were sectioned at 60 nm using an ultramicrotome, and the sections were viewed using a Tecnai G2 transmission electron microscope at 120 kV. Confocal micrographs were produced as high-quality images using PhotoZoom Pro 8 software. Enlarged fluorescence images were fitted to the electron micrographs using the Image J BigWarp program.

Immunocytochemistry
The transfected cells were washed with PBS, fixed with 4% PFA for 10 min, and permeabilized with 0.1% Triton X-100 for 10 min. Then, they were blocked with 3% bovine serum albumin for 1 h at room temperature prior to incubation overnight with anti-Myc) or anti-FLAG antibodies at 4℃ and then with fluorescent conjugated anti-mouse secondary antibodies for 2 h at room temperature. Finally, the cells were washed three times with 1× PBS and mounted onto glass slides. The preparations were analyzed using an LSM 880 confocal laser scanning microscope.

Plasmid Constructions for Crystallization and ITC
Human TP53INP2 LIR (residues 28-40) and fission yeast Hfl1 LIR (residues 386-409) were fused to the N-terminus of human GABARAP and GABARAP-L2 with F3S/V4T and W3S/M4T mutations, respectively, for the promotion of crystallization. All genes were inserted downstream of the sequence encoding the human rhinovirus (HRV) 3C protease recognition site in the pGEX6P-1 vector, except for the Hfl1 LIR-GABARAP-L2 fusion protein, which was inserted upstream of the sequence encoding the HRV 3C protease site, following the myelin basic protein (MBP) gene of the pET15b-MBP vector. The plasmids were constructed using NEBuilder HiFi DNA Assembly Master Mix or an In-Fusion HD Cloning Kit.

Protein and Peptide Purification
All proteins were expressed in E. coli BL21 (DE3). For protein purification, the bacteria were cultured at 37 °C to an OD600 of 0.8-1.0 and then supplemented with IPTG at 100 µM and further incubated overnight at 16 °C. The bacterial pellets were resuspended in PBS and 5 mM EDTA and sonicated for 10 min. After centrifugation, the supernatants were recovered and subjected to GST-accept resin. The resin was washed three times with PBS and eluted with 10 mM glutathione and 50 mM Tris at pH 8.0. The eluates were desalted with PBS using a Bio-Scale Mini Bio-Gel P-6 Desalting Column and then digested overnight with HRV 3C protease at 4 °C. Artificial glycine-proline sequences were retained at the N-terminus of the gene product, except for the Hfl1 LIR-GABARAP-L2 fusion protein, which retained artificial L-E-V-L-F-Q sequences at the C-terminus. All mutations were generated via PCR-based mutagenesis. The samples were subjected to GST-accept resin to remove the digested GST tags, and the flow-through fractions were recovered. The peptides were synthesized by Cosmo Bio.
Briefly, the peptides (10 mg) were dissolved in 300 µl distilled water and ~10 µl ammonium hydrate. The proteins and peptides were purified by size-exclusion chromatography with 20 mM HEPES at pH 6.8 and 150 mM sodium chloride using a Superdex 200 prep grade column or Superdex 75 10/300 GL column.

Crystallization and Diffraction Data Collection
All crystallizations were performed at 20 °C using the sitting-drop vapor-diffusion method by mixing protein and reservoir solutions at a 1:1 volume ratio. For crystallization of TP53INP2 LIR-GABARAP fusion protein, 11.79 mg/ml protein was mixed with 10% 2-propanol, 0.1 M sodium phosphate/citric acid at pH 4.2, and 0.2 M lithium sulfate. For crystallization of SpHfl1 LIR-GABARAP-L2 fusion protein, 40.931 mg/ml protein was mixed with 8% polyethylene glycol 3000 and 0.1 M sodium citrate at pH 5.8. The crystals were soaked in cryoprotectant and frozen in liquid nitrogen. The cryoprotectants for SpHfl1 LIR-GABARAP-L2 and TP53INP2 LIR-GABARAP were prepared by supplementing each reservoir solution with 25% 2-methyl-2,4-pentanediol or 33% glycerol, respectively. The flash-cooled crystals were maintained under nitrogen gas at 178 °C during data collection. Diffraction data were collected using an EIGER X4M detector attached to the beamline BL-1A with a wavelength of 1.1000 Å. The diffraction data were indexed, integrated, and scaled using energy-dispersive X-ray spectroscopy 43 .

Structure Determination
The structures of the SpHfl1 LIR-GABARAP-L2 and TP53INP2 LIR-GABARAP fusion proteins were determined by the molecular replacement method using the Phenix program 44 .
GABARAP (PDBID: 1GNU) and GABARAP-L2 (PDBID: 4CO7) structures were used as the search model. Crystallographic refinement was performed using the Phenix and Coot programs 44,45 . All structural images were prepared with PyMOL Molecular Graphics System v2.0.

Isothermal Titration Calorimetry
ITC experiments were performed using a MicroCal iTC200 calorimeter at 25 °C with stirring at 1,000 rpm. SpHfl1 and TP53INP2 peptides were prepared at 500 µM as injection samples.
LC3 family and GABARAP subfamily proteins were prepared at 50 µM as cell samples. After a test injection of 0.4 µl, titration involved 18 injections of 2 µl of injection samples at intervals of 120 s into the cell. The datasets obtained from titration of the peptides into cells filled with buffer were used as reference data to subtract the heat of dilution. MicroCal Origin 7.0 software was used for data analysis. The thermal measurement data for the first test injection of the syringe samples were removed from the analysis. The thermal titration data were fitted to a single-site binding model that was used to determine thermodynamic parameters, such as enthalpy, Kd, and stoichiometry of binding (N). When the fit was not convergent due to weak interactions, N was fixed at 1.0. The error for each parameter represented the fitting error.

Quantification and Statistical Analysis
All data were presented as mean + SEM and performed in triplicate or higher. The Kolmogorov-Smirnov normality test was performed to check the Gaussian distribution of the group. For multiple group comparison, one-way ANOVA in conjunction with the Newman-Keuls multiple comparison test or the Kruskal-Wallis test followed by a Dunn multiple comparison test was performed as a parametric or non-parametric test, respectively. Statistical analysis was accomplished using GraphPad Prism 6. P values <0.05 were considered statistically significant.