Development of new tools to study membrane-anchored mammalian Atg8 proteins

ABSTRACT Mammals conserve multiple mammalian Atg8-family proteins (mATG8s) consisting of GABARAP (GABA type A receptor-associated protein) and MAP1LC3/LC3 (microtubule associated protein 1 light chain 3) subfamilies that tightly bind to autophagic membranes in a membrane-anchored form. These proteins are crucial for selective autophagy and recruit proteins bearing LC3-interacting region (LIR) motifs. However, because limited research tools are available, information on the specific roles of each membrane-anchored mATG8 in selective autophagy is scarce. In this study, we identified LIR motifs specific to the membrane-anchored form of each mATG8 and characterized the residues critical for their selective interaction using cell-based assays and structural analyses. We then used these selective LIR motifs to develop probes and irreversible deconjugases that targeted selective membrane-anchored mATG8s in the autophagic membrane, revealing that membrane-anchored GABARAP subfamily proteins regulate the aggrephagy of amyotrophic lateral sclerosis-linked protein aggregates. Our tools will be useful for elucidating the functional significance of each mATG8 protein on autophagic membranes in autophagy research. Abbreviations A:C autophagic membrane:cytosol; ALS amyotrophic lateral sclerosis; ATG4 autophagy related 4; Atg8 autophagy related 8; BafA1 bafilomycin A1; BNIP3L/Nix BCL2 interacting protein 3 like; CALCOCO2/NDP52 calcium binding and coiled-coil domain 2; EBSS Earle’s balanced salt solution; GABARAP GABA type A receptor-associated protein; GST glutathione S transferase; HKO hexa knockout; Kd dissociation constant; LIR LC3-interacting region; MAP1LC3/LC3 microtubule associated protein 1 light chain 3; NLS nuclear localization signal/sequence; PE phosphatidylethanolamine; SpHfl1 Schizosaccharomyces pombeorganic solute transmembrane transporter; SQSTM1/p62 SQSTM1/p62; TARDBP/TDP-43 TAR DNA binding protein; TKO triple knockout


Introduction
Macroautophagy/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].
Autophagy is tightly regulated by several ATG (autophagy related) proteins. To date, more than 40 ATG genes have been identified in yeast and more complex 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 ubiquitin-like protein in yeast that is covalently conjugated to phosphatidylethanolamine (PE) in phagophores following proteolytic cleavage of its C terminus by the cysteine protease ATG4. ATG4 is also involved in the delipidation of ATG8-PE to release Atg8 from autophagosomes. This cleavage is necessary for the expansion and closure of the phagophore membrane. Although there is only one ATG8 protein in yeast, mammals have two subgroups: the family of MAP1LC3/LC3 (microtubule associated protein 1 light chain 3) proteins (LC3A, LC3B, LC3B2 and LC3C) and GABARAP (GABA type A receptor-associated protein) proteins (GABARAP, GABARAPL1, and GABARAPL2) [6]. These pro-teins undergo reversible lipidation via conjugation of PE to their C-terminal regions on the phagophore. Although mammalian ATG8 protein (mATG8) conjugation is crucial for conventional autophagy processes, including autophagosome biogenesis and maturation, accumulating evidence indicates 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 non-conventional autophagy or even in autophagy-independent pathways is largely unknown.
mATG8 proteins recruit autophagic machinery containing LC3-interacting region (LIR) motifs into phagophores [8,9]. They can also sequester selective cargo into phagophores via LIR motif-containing receptors during selective autophagy [10]. Therefore, the specific roles of LC3 and GABARAP subfamily proteins are regulated by several LC3-and GABARAP-interacting proteins with LIR motifs. Many mATG8-interacting proteins contain canonical LIR motifs that have a basic hydrophobic LIR motif with a core consensus sequence of (W/F/Y)-X-X-(L/I/V), which binds to the W-and L-sites conserved in mATG8 proteins using the amino acids W/F/Y and L/I/V, respectively [8,[11][12][13][14]. The GABARAP-interacting motif (GIM) was recently proposed to have a core consensus sequence of (W/F)-(I/V)-X-V, similar to that of a 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 involved in mATG8 interactions [16,17]. Many LIR motifs that interact with mATG8 have been extensively investigated in recent studies [9,18,19]. However, the mechanism underlying selective binding remains largely unknown. Several studies have used peptide-based arrays, glutathione S transferase (GST) affinity-isolation assays, or competitive time-resolved Förster/fluorescence resonance energy transfer to examine the binding properties of LIR motifs to mATG8 [9,18,[20][21][22]. Although these approaches are useful, the assays are purely in vitro and may only reflect the binding properties of the LIR motifs of soluble cytosolic mATG8 proteins. Thus, many in vitro assay results may not reflect the physiological binding properties of membrane-anchored mATG8 proteins on autophagic membranes in cells. Therefore, identification of functional LIR motifs and characterization of the determinants that result in selective binding of membrane-anchored mATG8 proteins to autophagic membranes in cells is crucial.
In this study, we used novel cellular assays to determine the binding properties of LIR motifs to cytosolic and membrane-anchored mATG8 proteins. We used structural analyses to identify LC3C-, GABARAPL2-, and GABARAP and GABARAPL1-selective LIR motifs and characterized the biochemical properties responsible for their selective binding. Furthermore, we generated a new system that selectively monitors or delipidates mATG8s anchored on the autophagic membrane by incorporating the 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 GABARAP subfamily proteins regulate the cellular degradation of amyotrophic lateral sclerosis (ALS)linked TDP-25 (the 25-kDa C-terminal fragment of TARDBP/ TDP-43) protein aggregates during aggrephagy. These data demonstrated that our newly developed tools can be widely applied to elucidate the functional significance of membraneanchored forms of each mATG8 protein in diverse autophagic and non-autophagic processes.

Characterization of binding properties of LIR motifs toward cytosolic and membrane-anchored mATG8s in hexa mATG8-knockout HeLa (HKO) cells
To examine the binding properties of LIR motifs fo each cytosolic mATG8 and to identify selective mATG8-binding LIR motifs in cells, we first used a LIR motif tagged with a 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 were not expressed as described previously [23] ( Figure S1; Figure 1A). If a LIR motif interacts with a green fluorescent protein (GFP)-tagged mATG8 in the cytosol, the LIR-mRFP-3xNLS sequesters cytosolic mATG8 in the nucleus, depending on the binding preference of the LIR-mRFP-3xNLS protein in HKO cells ( Figure 1A). This enabled the quantification of the relative GFP fluorescence intensity ratio between the nucleus and cytosol (N:C ratio) in live cells.
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), leading to inhibition of cellular localization to phagophores and consequently autophagosomes. We first examined the binding properties of a LIR motif from SQSTM1/p62 (LIR [SQSTM1]), the first LIR motif for mATG8 proteins identified in the autophagic membrane [24]. Each GFP-mATG8(GA) protein was primarily localized to the nucleus in cells expressing LIR(SQSTM1)-mRFP-3xNLS but not in cells expressing LIR motif mutant LIR(SQSTM1) m -mRFP-3xNLS, a LIR motif mutant, in which two amino acids in the core sequence of LIR (SQSTM1) were mutated to alanine or mRFP-3xNLS ( Figure 1B). The relative N:C ratio in LIR(SQSTM1)-mRFP -3xNLS-expressing cells was significantly higher than that in LIR(SQSTM1) m -mRFP-3xNLS-or mRFP-3xNLS-expressing cells for all mATG8s, indicating that this LIR bound to all types of cytosolic mATG8 in cells in a LIR-dependent manner ( Figures 1B,C). Consistent with the N:C ratio assay, LIR (SQSTM1)-GFP, but not LIR(SQSTM1) m -GFP, bound to all mATG8 proteins in GST pull-down and coimmunoprecipitation (co-IP) experiments using GST-mATG8 or 3xFlag-mATG8 proteins, respectively ( Figures 1D,E), confirming that LIR(SQSTM1) bound to all cytosolic mATG8 proteins.
Next, we determined the binding properties of LIR motifs to autophagic membrane-anchored mATG8s. Thus, the translocation of the LIR motif from the cytosol to the mRFP-mATG8-containing autophagic membrane during their interaction was 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 a LIR motif for each expressed membrane-anchored mRFP-mATG8 can be determined Figure 1. Development of new methods to detect the binding affinities of LIR motifs for cytosolic or membrane-anchored mATG8 proteins in cells. (A) Schematic model for the detection of binding affinities of LIR motifs toward soluble/cytosolic mATG8 proteins in HKO cells using LIR-mRFP-3xNLS and GFP-mATG8(GA) constructs. If a LIR motif interacts with a GFP-tagged mATG8 in the cytosol, the LIR-mRFP-3xNLS should sequester the cytosolic mATG8 in the nucleus (right). If it does not, or weakly interacts with GFP-tagged mATG8(GA), the distribution of GFP-mATG8(GA) will not be changed (left). LIR; LC3-interacting region motif, NLS; nuclear localization signal/sequence, mATG8(GA); mammalian Atg8 mutants in which the C-terminal glycine residue was replaced with alanine to impair PE conjugation (lipidation), LIR-R-3xNLS; LIR(x)-mRFP-3xNLS, G-mATG8 GA ; GFP-mATG8(GA). (B-C) the LIR motif from SQSTM1 (LIR[SQSTM1]) interacts with all cytosolic mATG8 proteins. (B) Cellular localization of LIR(SQSTM1)-mRFP-3xNLS and LIR(SQSTM1) m -mRFP-3xNLS, a LIR motif mutant, in which two amino acids in the core sequence of LIR(SQSTM1) were mutated to alanine or mRFP-3xNLS in HKO cells. (C) Quantification of the nuclear and cytosolic (N:C) ratio of GFP fluorescence. *** P < 0.001, one-way ANOVA in conjunction with the Newman -Keuls multiple comparison test. Values are presented as means + SEM (n ≥20 for each group). (D-E) Glutathione S transferase (GST) affinity-isolation assays (D) and 3xFlag co-immunoprecipitation (co-IP) experiments (E) to analyze the binding of LIR(SQSTM1)-GFP and LIR(SQSTM1) m -GFP. the co-precipitated LIR(SQSTM1)-GFP proteins (upper panel) were analyzed by western blot using the indicated anti-GFP antibodies. The immobilized GST fusion proteins used are displayed on coomassie brilliant blue (CBB)-stained gels (lower panel). Data from one representative experiment of three independent experiments is presented. (F) Schematic model of the detection of binding affinities of LIR motifs toward membrane-anchored mATG8 proteins in HKO cells using LIR(x)-GFP and mRFP-mATG8 constructs. If a LIR motif interacts with an autophagic membrane-anchored mRFP-mATG8, the localization of LIR(X)-GFP will ( Figure 1F). To maximize visual autophagosomes in cells overexpressing each mRFP-mATG8(GA) and either LIR (SQSTM1)-GFP, LIR(SQSTM1) m -GFP, or GFP, we treated HKO cells with 100 nM bafilomycin A 1 (BafA1) blocking for autophagosome-lysosome fusion under starvation induction (Earle's balanced salt solution [EBSS]) for 2 h. As shown in Figure S2, an mRFP-tagged LC3B protein exclusively localized to autophagosomes, whereas each mRFP-mATG8 (GA) protein diffusely localized mainly to the cytosol and nucleus in HKO cells treated with 100 nM BafA1 under starvation induction for 2 h, indicating that unlipidated mATG8s are expressed diffusely in cells. To validate the binding of the SQSTM1 LIR motif to mATG8 proteins anchored to the autophagic membrane, we expressed LIR (SQSTM1)-GFP, LIR(SQSTM1) m -GFP, or GFP with each mRFP-mATG8 in HKO cells treated with 100 nM BafA1 in EBSS for 2 h. We then quantified the relative GFP fluorescence intensity ratio between the autophagosome and cytosol (A:C ratio) ( Figures 1G,H). As expected, the relative A:C ratio in LIR(SQSTM1)-GFP-expressing cells was significantly higher than that in LIR(SQSTM1) m -GFP-or GFP-expressing cells for the LC3 subfamily. Unexpectedly, the relative A:C ratio in LIR(SQSTM1)-GFP-expressing cells was not significantly different from that in LIR(SQSTM1) m -GFP-or GFPexpressing cells for the GABARAP subfamily ( Figures 1G,H). Our results indicated that LIR(SQSTM1)-GFP preferentially localized to membrane-anchored LC3 subfamily proteins, but not to membrane-anchored GABARAP subfamily proteins in cells. Taken together, our data raise the possibility that LIR (SQSTM1) could interact with all types of cytosolic mATG8s but was targeted to the autophagic membrane in cells, probably via preferential binding to LC3 subfamily proteins anchored to the autophagic membrane in cells. Consistent with this, it has been reported that SQSTM1 binds to both soluble LC3B and GABARAPL2 in the cytosol but only to autophagic membrane-anchored LC3B [25]. Therefore, our newly developed methods can identify the binding properties of LIR motifs for cytosolic or membrane-anchored mATG8s in cells, respectively.

Identification of LIR motifs that bind to autophagic membrane-anchored mATG8
As our goal was to generate a new system that selectively monitors or delipidates autophagic membrane-anchored mATG8s, the A:C ratio assay was a useful strategy for identifying LIR motifs that selectively bind to autophagic membrane-anchored mATG8s. Therefore, we examined the binding properties of various GFP-fused, known LIR motifs with mRFP-mATG8 proteins in HKO cells using the A:C ratio assay (Table 1; Figure S3; Figure 2). To this end, we expressed each LIR(X)-GFP (X: the protein from which the LIR motif originated) together with each mRFP-mATG8 in HKO cells treated with 100 nM BafA1 in EBSS for 2 h. We then quantified the A:C ratio of each mATG8. Finally, the value was normalized to that of GFP and each mRFP-mATG8 -expressing cell, as shown in Figures 1G,H (the normalized A: C ratio).
We first identified LIR motifs that preferentially bind to GABARAPL2 (Figures 2A,B). An atypical LIR motif from SpHfl1 (LIR[Sp]), which binds to Atg8 in yeast [26], was mostly localized to GABARAPL2-and weakly but significantly localized to GABARAP-and GABARAPL1-containing autophagic membranes (Figure 2A). Intriguingly, LIR(Sp)-GFP exhibited strong binding to cytosolic GABARAPL2 and weak binding to GABARAPL1 in a GST pull-down assay ( Figure 2B). Isothermal titration calorimetry (ITC) experiments further confirmed the strong and specific binding affinity of LIR(Sp) to GABARAPL2; the dissociation constant (K d ) of LIR(Sp) to GABARAPL2 was 0.24 μM, which was 8-12 times lower than that to GABARAP (1.8 μM) and GABARAPL1 (2.9 μM) ( Figure S4A). The corresponding values for the LC3 subfamily proteins could not be determined because the interactions were too weak ( Figure S4A). Thus, LIR(Sp) preferentially binds cytosolic and membraneanchored GABARAPL2.
Next, we identified the LIR motif that preferentially localized to GABARAPL1-containing autophagic membranes ( Figures 2C,D). The LIR motif from BNIP3L/Nix (LIR [BNIP3L]) exhibited no significant binding to any mATG8 (Table 1 and Figure S3). However, a phosphomimetic mutant of BNIP3L (LIR[BNIP3L-pm]), in which two serine residues located at the N terminus of the core LIR motif were substituted by phosphomimetic E residues, exhibited selective localization to GABARAPL1-containing autophagic membranes [27] ( Figure 2C). In contrast, in the GST affinityisolation assay, LIR(BNIP3L-pm) interacted with all the mATG8 proteins ( Figure 2D). Intriguingly, these results suggest the possibility that LIR(BNIP3L-pm) binds to all cytosolic mATG8 proteins but selectively binds to autophagic membrane-anchored GABARAPL1 in cells.
Additionally, we identified the LIR motif that preferentially localizes to GABARAP-and GABARAPL1-containing autophagic membranes ( Figures 2E,F). The LIR motif from the C terminus of ATG4B (LIR[4B]) is bound to all autophagic membrane-anchored mATG8 proteins, except for LC3C (Table 1 and Figure S3). A previous two-dimensional peptide array scan 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 [28]. Consistent with this report, be shifted from the cytosol to the mRFP-mATG8-containing autophagic membrane in cells (right). If the interaction does not occur or is weak, the localization of LIR(X)-GFP will be not changed (left).  bound to all membrane-anchored mATG8 proteins (Table 1). Meanwhile, the W to T mutation within the core region of LIR(TP), referred to as LIR(TP[T]), which makes this LIR motif atypical, induced selective binding to membrane-anchored LC3C ( Figure 2G). Our GST pull-down assay confirmed the selective binding of LIR(TP[T])-GFP to LC3C ( Figure 2H). ITC experiments also demonstrated the selectivity of LIR(TP[T]) toward LC3C ( Figure S4B). The K d value of LIR(TP[T]) binding to LC3C was 26.9 μM, five times lower than that of the other mATG8 proteins (>132 μM), indicating that LIR (TP[T]) could be used as an LC3C-selective LIR motif ( Figure  S4B). Taken together, we successfully identified selective LIR motifs for membrane-anchored forms of LC3C, GABARAP and GABARAPL1, GABARAPL1, and GABARAPL2.

Structural basis for selective binding of mATG8-binding LIR motifs
We attempted to determine the crystal structures of atypical LIRs bound to cytosolic mATG8 proteins to reveal the specific mechanism underlying the interaction between LIR and mATG8 proteins at the atomic level. We successfully determined the structure of the LIR(Sp)-GABARAPL2 fusion protein at a resolution of 1.86 Å ( Figure 3A; Table S1). The structure of the LIR(Sp)-GABARAPL2 complex is similar to that of the LIR(Sp)-SpAtg8 complex ( Figure S5A). The SpHfl1 LIR consists of an α-helix, from D391 to M404, and an N-terminal tail. The helix formed extensive hydrophobic interactions with V51, P52, I55, W62, and I63 of GABARAPL2 using M394, L397, Y398, A401, and M404. Among these, Y398 formed the most critical interaction, as it was inserted deep into the L-site pocket of GABARAPL2. In addition to hydrophobic interactions, three acidic residues of LIR(Sp), D391, E393, and E395, formed electrostatic interactions with K46, R67, and R28 in GABARAPL2, respectively ( Figure 3B) [26]. Although most of the interactions were similar to those in the LIR(Sp)-SpATG8 structure, F388 of SpHfl1 was not inserted into the W-site of GABARAPL2. Considering that the alanine substitution of F388 induced a limited decrease in the binding affinity to SpATG8, this observation implies that W-site binding is not important for the binding of SpHfl1 to Atg8-family proteins. We also performed sequence alignment of mATG8 proteins ( Figure S5B) [26]. Among the GABARAPL2 residues involved in the interaction with LIR(Sp), W62 was the sole residue that was not conserved in other mATG8 proteins (F, K, or S was also observed at this position). To assess the importance of W62 in this interaction, we performed ITC experiments using three mATG8 mutants (GABARAPL2 W62A, GABARAP F62W, and LC3B K65W) ( Figure 3E). The W62A mutation in GABARAPL2 reduced its binding affinity to LIR(Sp), but the F62W mutation in GABARAP had the opposite effect. K65W mutation in LC3B marginally increased this interaction. Coupled with the fact that SpAtg8, which binds strongly to SpHfl1, has a Y at position 62, these data suggest that either W or Y at position 62 is necessary, but not sufficient, for the strong binding of ATG8 family proteins to SpHfl1 LIR.
The crystal structure of the LIR(TP)-mATG8 complex has not been reported previously. We failed to crystallize the LIR(TP) mutants bound to mATG8 proteins but succeeded in crystallizing and determining the structure of the wild-type (WT) LIR(TP)-GABARAP fusion protein at a resolution of Underline; core region in LIR motif. The normalized A:C ratio: −1.4: -; 1.4-2.0: +; 2.0-4.0: ++; 4.0: +++ AS_67 is a LC3C-specific peptide originated from a previous report [15]. 1.75 Å ( Figure 3C; Table S1). The conformation of LIR(TP) was unique compared to that of canonical LIRs, where 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 [29]. Although the LIR(TP) peptide used for crystallization possesses six acidic residues, only one residue (E29) forms an electrostatic interaction with GABARAP (R67), suggesting that the binding affinity is largely dependent on the core LIR motif. To assess the specificity of LIR(TP[T]) to LC3C, we prepared a structural model of the LIR(TP[T])-LC3C complex by superimposing the WT LIR(TP)-GABARAP structure onto the LC3C-CALCOCO2/NDP52 structure (Protein Database ID [PDBID]:3VVW) followed by manual model adjustment ( Figure 3D) [16]. The CALCOCO2 LIR possessed a noncanonical 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 W-site. The lack of canonical interactions must be compensated by additional interactions to maintain a 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 formed hydrophobic and electrostatic interactions with LC3C F33 and K32, respectively. Additionally, E29 and E31 of LIR(TP[T]) form electrostatic interactions with R76 and K55 of LC3C, respectively. Among these residues, K55 and R76 were strictly conserved in all mATG8 proteins, K32 was conserved within the GABARAP subfamily, and 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 S6A). We noticed that the α2 helix of the LC3 family, whose C terminus contained K32 and F33, was located closer to the LIR-binding pocket than that of GABARAP subfamily proteins ( Figure S5C, 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 amino acid at position 18 (using GABARAP numbering). The LC3 family possesses a V at this position, which has a larger side chain than G (GABARAP and GABARAPL1) and S (GABARAPL2), resulting in the steric crush of the α2 helix with the ubiquitin fold, thereby positioning the α2 helix toward the LIR-binding site ( Figure S5C, right). Consistent with this, the mutation of V26 to G (corresponding to GABARAP and GABARAPL1) reduced the association between the LC3C-positive autophagic membrane and LIR (TP[T]) ( Figure S6A). Therefore, the combination of K32, F33, and the properly positioned α2 helix is likely to be necessary for the specific binding of LC3C to LIR(TP[T]).

Development of probes that selectively monitor membrane-anchored mATG8 proteins on autophagosomes
Next, we monitored membrane-anchored mATG8-containing autophagic membranes to study the function of each mATG8 protein in autophagy or selective autophagy, using each identified selective 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 mATG8-binding LIR motifs (gProbe-X, X: the protein name from which the LIR motif originated), as described previously [30]. Each probe, which contained two identical LIR motifs and an MT domain, was co-expressed with each of the mRFP-mATG8 proteins in HKO cells ( Figure 4A). At 24 h after transfection, the normalized A:C ratio was quantified in the HKO cells treated with 100 nM BafA1 in EBSS for 2 h (Table S2 and Figures 4B-E). Among the selective LIR motifs that we characterized, LIR(Sp) was selective for GABARAPL2 (Figures 2A,B). Therefore, we first tested whether gProbe-Sp selectively localized to GABARAPL2-containing autophagic membranes. Indeed, gProbe-Sp was mostly localized to GABARAPL2containing autophagic membranes; however, it was also localized to other types of mATG8-containing autophagic membranes, indicating that gProbe-Sp can detect all types of mATG8-containing autophagic membranes at different levels ( Figure 4B and Table S2). L397 in the LIR(Sp) hydrophobically interacted with W62 in GABARAPL2, which corresponds to F62 in GABARAP and GABARAPL1 ( Figure 3B). Therefore, we considered L397 a good candidate mutation to generate a more selective GABARAPL2-binding LIR motif. We replaced L397 with I (LIR(Sp[I]) to generate gProbe-Sp (I), which selectively localized to GABARAPL2-containing autophagic membranes (Figures 4C and Table S2), suggesting that gProbe-Sp(I) is a highly selective probe for GABARAPL2containing autophagic membranes.
To further examine whether our selective LIR-based sensors could indeed detect autophagosomes at the ultrastructural level, we performed correlative light and electron microscopy (CLEM) analysis to probe positive spots of interest in whole-cell images with fluorescence light microscopy and then zoomed in for a closer look using electron microscopy. This dual examination provided valuable, complementary, and unique information regarding the cellular and ultrastructural components of endogenous mATG8containing autophagosomes. To this end, rProbe-Fy (GFP in gProbe-Fy replaced by mRFP, r: mRFP) and gProbe-Sp(I) were co-transfected into WT HeLa cells to determine whether membrane-anchored LC3 subfamily-or GABARAPL2selective probes detected the same or distinct autophagosomes. HeLa cells were treated with rapamycin (100 nM) in the presence of BafA1 (100 nM) for 2 h to maximize the number of autophagosomes. Both rProbe-Fy and gProbe-Sp (I) detected autophagosomes throughout the cell (Figures 4F-J). Notably, some autophagosomes were detected only by red signal, indicating only rProbe-Fy-containing autophagosome ( Figure 4J). Taken together, these data suggest that our new selective LIR-based probes are useful for identifying autophagosomes containing distinct mATG8 subfamilies at cellular and ultrastructural levels.  Figures 1G,H (the normalized A:C ratio). Values are presented as means + SEM (n ≥20 for each group). ***P < 0.001 compared with all other mRFP-mATG8 expressing groups with one-way ANOVA in conjunction with the Newman -Keuls multiple

Development of enzymes that selectively delipidate mATG8-PE in autophagic membranes
To date, due to limitations in tools that can selectively inhibit or deplete each membrane-anchored mATG8 protein, information concerning the specific roles of the cytosolic and membraneanchored forms of each mATG8 protein in autophagy and autophagy-independent pathways is scarce. Compared to 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 [31,32]. Therefore, to generate deconjugases that selectively remove PE from mATG8-PE in our study, we replaced LIR1/2 and LIR3 in 3xFlag-fused catalytically active RavZ protein (deconjugase) with two identical LIR(  (Figures 5B,C). Likewise, the expression of deconjugase-St, deconjugase-TP(T), deconjugase-4B(T), and deconjugase-Sp(I) selectively reduced GFP-GABARAP subfamily-containing, GFP-LC3C-containing, GFP-GABARAP-and GABARAPL1-containing, and GFP-GABARAPL2-containing autophagic membranes, respectively.
We detected endogenous levels of membrane-anchored mATG8 proteins in deconjugase-X (X: Fy, St, TP[T], 4B[T], or Sp[I])-expressing HEK293T cells treated with 50 μM chloroquine (CQ) impairing autophagosome-lysosome fusion for 4 h to increase mATG8-II level using western blot analysis to further determine enzyme specificity. Since the expression level of endogenous LC3C was too low to be detected by the anti-LC3C antibody, we expressed 3xFlag-LC3C and detected this using an anti-Flag antibody. Furthermore, we expressed deconjugases using a lentiviral vector to improve the expression levels. The expression of deconjugase-Fy strongly reduced the level of membrane-anchored LC3A and LC3B (LC3A-II and LC3B-II), but also significantly reduced that of membrane-anchored LC3C (LC3C-II) and membrane-anchored GABARAP subfamily (GABARAP-II, GABARAPL1-II and GABARAPL2-II) proteins, whereas deconjugase-St induced a selective reduction in the membrane-anchored form of GABARAP subfamily proteins, indicating selective delipidation (Figures 5D,E). Additionally, deconjugase-Sp(I) selectively downregulated membraneanchored GABARAPL2 (GABARAPL2-II). In contrast, deconjugase-TP(T) and deconjugase-4B(T) non-selectively reduced membrane-anchored mATG8s ( Figure S7). These results indicate that deconjugase-St and deconjugase-Sp(I) could be used to selectively delipidate the GABARAP subfamily and GABARAPL2, respectively. In the case of deconjugase-Fy, further improvement is required to selectively deplete the LC3 subfamily.

Membrane-anchored GABARAP subfamily proteins regulate TDP-25-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 human diseases, including neurodegenerative diseases, and may be cleared via aggrephagy, little is known about the differential roles of membrane-anchored LC3 or GABARAP subfamily proteins in aggrephagy.
Moreover, we examined the size and number of TDP-25 aggregates in neurons expressing deconjugase, deconjugase-Fy, deconjugase-St, or deconjugase-Sp(I) after autophagy induction with trehalose in the presence or absence of BafA1 (100 nM, 2 h), to distinguish their differential roles in the regulation of TDP-25 aggregates by aggrephagy. We found that deconjugase expression failed to reduce the size of TDP-25 aggregates, indicating that deconjugase cleaved and inhibited all membrane-anchored mATG8 proteins, leading to cellular defects in aggrephagy (Figures 6A-C). Interestingly, deconjugase-St and deconjugase-Sp(I)-expressing neurons failed to reduce the size of TDP-25 aggregates compared to deconjugase-Fy-expressing neurons ( Figure 6A-C). These data indicate that the membrane-anchored form of GABARAP subfamily proteins, but not LC3 subfamily proteins, regulates the degradation of TDP-25 aggregates during aggrephagy.
To further confirm whether the GABARAP subfamily regulates cellular degradation of TDP-25 aggregates during aggrephagy, we expressed MYC-TDP-25 in WT, LC3 triple knockout (TKO), and GABARAP TKO HeLa cells, and examined the size of TDP-25 aggregates after autophagy induction with trehalose in the presence or absence of BafA1 (100 nM) for 2 h. As shown in Figures 7A-D, the size of TDP-25 aggregates was significantly reduced by autophagy induction in WT and LC3 TKO cells, but not in GABARAP TKO cells. This reduction is mediated by autolysosomes, because the inhibition of autophagy and lysosome fusion completely blocked this reduction (Figure 7). Taken together, our data strongly suggest that membrane-anchored GABARAP subfamily proteins, but not LC3 subfamily proteins, are involved in the cellular degradation of TDP-25 aggregates during aggrephagy.

Discussion
LIR-based sensors have been developed to detect endogenous LC3/GABARAP subfamily proteins [15,35,36]. Most binding experiments with ATG8 proteins have been performed with cytosolic unlipidated/free mATG8 proteins. However, the mATG8 proteins frequently perform their function on autophagic membranes in the membrane-anchored/lipidated form. There is limited knowledge regarding the binding of mATG8s to autophagic membranes. In this study, we used cellular assays to determine cytosolic and autophagic membrane-anchored mATG8 proteins binding in cells. We improved the previous version of LIR-based autophagosome sensors by introducing highly selective LIR motifs for LC3C, GABARAP and GABARAPL1, and GABARAPL2. These improved sensors can detect different sets of autophagic membrane-anchored mATG8 proteins of live cells. For example, gProbe-TP(T) detects LC3Ccontaining autophagic membranes in WT HeLa cells but not in ATG5 or ATG7 KO HeLa cells or MEF cells, in which LC3C is not expressed (Figure S6B), thereby monitoring LC3Ccontaining autophagic membranes in live cells. Therefore, our LIR-based LC3-or GABARAP subfamily selective sensors will likely provide novel insights into the differential functions of mATG8 proteins by monitoring their cellular localization on autophagic membranes at different stages of macroautophagy/ selective autophagy or nonconventional autophagy associated with membrane-anchored LC3/GABARAP subfamily proteins.
Although the specific roles of each mATG8 protein in selective autophagy are largely unknown, ATG8 is recruited together with cargo into autophagosomal membranes via interactions with autophagy receptors to facilitate cargo degradation [37]. Most autophagy receptors possess a LIR motif that allows their direct binding to LC3, whereas most autophagy receptor proteins have GABARAP-specific LIRs [38]. Therefore, it has been proposed that LC3 subfamily proteins are essential for cargo recruitment during selective autophagy [7]. However, in these studies, the specific roles of the cytosolic and membrane-anchored forms of each mATG8 during macroautophagy/selective autophagy could not be distinguished. Many studies have used knockdown or knockout systems for ATG8 proteins or ATG8 conjugation-deficient cells to study the function of mATG8s that lead to the depletion of both soluble (cytosolic) and membrane-anchored mATG8 proteins or inhibit the lipidation of all mATG8 proteins. To our knowledge, this is the first report on the use of irreversible selective mATG8 deconjugases consisting of ATG8-selective LIR motifs and the catalytic domain of RavZ to investigate the functional significance of membrane-anchored mATG8 proteins in several types of autophagy. The use of selective deconjugases revealed that endogenous GABARAP subfamily protein-containing autophagosomes preferentially localize to and contact TDP-25 aggregates. More importantly, GABARAP proteins, including GABARAPL2 in autophagosomes (membrane-anchored GABARAP proteins), regulate the cellular degradation of TDP-25 aggregates in an autophagydependent manner ( Figure 6). Membrane-anchored LC3A/B may not be involved in the cellular degradation of TDP-25 aggregates, although some LC3A-and LC3Bpositive autophagosomes were associated with TDP-25 aggregates. The data we obtained using LC3 TKO or GABARAP TKO cells and our novel deconjugases strongly support our finding that membrane-anchored GABARAP family proteins regulate aggrephagy associated with TDP-25 aggregates (Figures 6 and 7). A study using mATG8knockout HeLa cells showed that GABARAP subfamily proteins regulate parkin-mediated mitophagy, a form of selective autophagy, whereas LC3 subfamily proteins mediate basal autophagy [23]. Harper et al. investigated the crucial roles of GABARAP subfamily proteins in selective autophagy, reporting that LC3 subfamily members were not involved in all steps of selective autophagy [39]. These results are consistent with our data, suggesting that membrane-anchored GABARAP subfamily proteins contribute to selective autophagy. Additionally, our cellular assays revealed that many LIR motifs selectively bind to GABARAP subfamily proteins and not to LC3A/B (Table 1). This might be because many different LIRcontaining autophagy receptors or other autophagy machinery proteins, including fusion components, selectively bind to GABARAP subfamily proteins to regulate their 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 membrane-anchored LC3-and GABARAP-selective LIR-based sensors and irreversible selective deconjugases for each autophagosome-bound mATG8-PE help elucidate the cellular localization and selective functions of these membrane-anchored mATG8 proteins in conventional and nonconventional autophagy.

Cell culture, transfection, confocal microscopy, and drug treatment
All cells used in the experiment were cultured in Dulbecco's modified Eagle's medium (Cytiva lifesciences, SH30262.01) supplemented with 10 or 15% (v:v) fetal bovine serum (Gibco, 26140079) and 1% (v:v) penicillin/streptomycin (Gibco, 15140122) in an incubator with 5% CO 2 at 37°C. Generation of Hexa mATG8 KO (HKO) HeLa, LC3 TKO and GABARAP TKO cells was performed as previously described [23]. All cells, including ATG5-or ATG7knockout HeLa cells, were seeded in a sticky-slide eightwell system (Ibidi, 80826) to obtain 40-60% confluent growth on the day of imaging. Cells were transfected with plasmid DNA constructs using calcium phosphate (Takara Bio, 631312) or Lipofectamine 2000 (Thermo Fisher Scientific, 11668027) 24 h before imaging. The relative amount of each construct was empirically determined based on the relative expression of each combination of constructs. HKO cells were treated with 100 nM bafilomycin A 1 (Sigma-Aldrich, B1793) in Earle's balanced salts solution (EBSS; Sigma-Aldrich, E3024) for 2 h to accumulate autophagosomes in cells. MEF was treated with 200 nM rapamycin (Sigma-Aldrich, R0395) in Dulbecco's modified Eagle's medium (DMEM) for 3 h to induce autophagy. The cells were observed under an inverted Zeiss LSM-700 scanning laser confocal microscope operated with the ZEN software. The laser excitation and spectral detection windows for fluorochromes were 488 nm (508-543 nm) for GFP and 561 nm (578-649 nm) for mRFP. Appropriate GFP (500-550 nm) and mRFP (575-625 nm) emission filters were used for the sequential imaging of each fluorescent protein.

Quantitative analysis of N:C or A:C ratios
To express the quantitative ratio of the nuclear:cytosol (N:C) fluorescence intensities, the average nuclear and cytosolic fluorescence intensities of at least five randomly selected points in the nucleus and cytosol in a single cell were measured using the ImageJ software. Similarly, the quantitative N: C ratio of at least 20 randomly selected cells was analyzed. The obtained values were normalized to the values calculated for mRFP-3xNLS-expressing cells (Figure 1). We also confirmed that the expression of mRFP-3xNLS lacking the LIR motif did not affect the N:C distribution of GFP-mATG8(GA). To calculate the ratio of autophagosome:cytosol (A:C) fluorescence intensities, autophagy flux was induced and blocked in HKO cells by treatment with 100 nM bafilomycin A 1 in Earle's balanced salt solution (EBSS) for 2 h. Cells were then fixed in 4% PFA (Cepham Life Sciences, 66311). The average value of autophagosome or cytosolic fluorescence intensity was obtained from at least five randomly selected points in autophagosomes or the cytosol of a single HKO cell using ImageJ. Similarly, 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 version 6.

Spot number analysis
To detect a reduction in mATG8-positive autophagosomes by enzyme activation of RavZ in autophagy-induced cells, we counted the number of spots over a certain field size in a single cell using ImageJ software. The cell image was changed to an 8-bit image and then inverted. The background was removed to ensure that only the spots remained visible. Finally, we counted the number of spots using the "Analyze the Particle" function of ImageJ. A minimum of 20 cells were quantified using this approach. All statistical data were calculated and plotted using GraphPad Prism version 6.

Co-immunoprecipitation (co-IP)
For transient transfection, HEK293T cells were plated at a density of 5-7 × 10 5 cells/well in six-well plates and cultured for 24 h. The cells were transfected with plasmid DNA sets each encoding LIR motif(x)-GFP and 3xFlag-mATG8s using calcium phosphate and incubated for 24 h. For Flag immunoprecipitation, the transfected HEK293T cells were washed with PBS, harvested, and lysed with EDTA lysis buffer solution (1% Triton X-100; 50 mM Tris-HCl, pH 7.5; 150 mM NaCl; 5 mM EDTA; protease inhibitor cocktail). The cell lysates were incubated overnight with 50 μl (bead volume) of mouse anti-Flag M2 antibodyconjugated beads (Sigma-Aldrich, A2220) at 4°C. The beads were subsequently washed three times with lysis buffer. The immunoprecipitates were eluted by adding 2 μg/ml of 3xFlag peptides (Sigma-Aldrich, F4799) and resuspended in SDS-PAGE sample buffer, boiled, and analyzed by SDS-PAGE.

Correlative light and electron microscopy
CLEM was performed as previously described [41]. Briefly, HeLa cells were cultured in culture dishes until 20-30% confluent and then transfected with mRFP-LC3A or -LC3B and EGFP-GABARAPL2 using Lipofectamine 2000. Next, the cells were treated with rapamycin (100 nM) in the presence of BafA1 (100 nM) for 2 h and imaged under a Ti-RCP confocal light microscope. The cells were fixed with 1% glutaraldehyde and 1% paraformaldehyde in 0.1 M cacodylate solution (pH 7.0) for 2 h at 4°C. Then, the fixed cells were washed with 0.1 M cacodylate solution, post-fixed with 2% osmium tetroxide for 1 h at 4°C and stained in 0.1 mg thiocarbohydrazide in 10 ml distilled water and en bloc in 1% uranyl acetate before dehydration through a graded ethanol series. Finally, the samples were embedded using the EMBed-812 embedding kit (Electron Microscopy Sciences, 14120). The embedded samples were sectioned at 60 nm using an ultramicrotome and viewed using a Tecnai G2 transmission electron microscope at 120 kV. Confocal micrographs were produced as high-quality images using the PhotoZoom Pro 8 software. The enlarged fluorescence images were fitted to electron micrographs using the ImageJ BigWarp program.

Immunocytochemistry
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 (Sigma-Aldrich, A2153) for 1 h at room temperature before incubation overnight with anti-MYC or anti-Flag antibodies (Sigma-Aldrich, F1804) at 4°C. Then, they were incubated in fluorescent conjugated anti-mouse secondary antibodies (Jackson ImmunoResearch Laboratories, 715-165-150) for 2 h at room temperature. Finally, the cells on the glass slides were washed twice with 1× PBS and the slides were mounted with mounting medium (Vectashield, H-1200). The preparations were analyzed using an LSM 880 confocal laser scanning microscope.

Plasmid constructions for crystallization and ITC
The human TP53INP2 LIR (residues 28-40) and fission yeast Hfl1 LIR (residues 386-409) were fused to the N terminus of human GABARAP and GABARAPL2 with F3S/V4T and W3S/M4T mutations, respectively, to promote crystallization. All genes were inserted downstream of the sequence encoding the human rhinovirus (HRV) 3C protease recognition site in the pGEX6P-1 vector (Cytiva, 28954648), except for the Hfl1 LIR-GABARAPL2 fusion protein, which was inserted upstream of the sequence encoding the HRV 3C protease site following the maltose-binding protein (MBP) gene of the pET15b vector (Merck, 69661-3CN). The plasmids were constructed using the NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs, E2621) or In-Fusion HD Cloning Kit (Takara Bio, 639648). The gene for HRV 3C protease was inserted between the GST gene and BamHI site of the pGEX6P-1 vector, from which the HRV 3C recognition site was removed.

Protein and peptide purification
All proteins were expressed in E. coli BL21 (DE3) cells. For protein purification, bacteria were cultured at 37°C until the OD 600 of the culture reached 0.8-1.0. Then, the culture was supplemented with 100 µM IPTG (Nacalai Tesque, 19742-07), and further incubated overnight at 16°C. The bacterial pellets were resuspended in PBS and 5 mM ethylenediaminetetraacetic acid (EDTA) and sonicated for 10 min. After centrifugation, the supernatants were recovered and subjected to a GSTaccepting resin (Nacalai Tesque, 09277-14). The resin was washed thrice with PBS and eluted with 10 mM glutathione (Nacalai Tesque, 17050-56) 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 (Bio-Rad Laboratories, 7325312) and then digested overnight with HRV 3C protease, which was purified by ourselves as described below, at 4°C. Artificial glycine-proline sequences were retained at the N terminus of the gene product, except for the Hfl1 LIR-GABARAPL2 fusion protein that retained artificial L-E-V-L-F-Q sequences at the C terminus. All mutations were generated using PCRbased mutagenesis. The samples were subjected to GST-accept resin to remove the digested GST tags and the flow-through fractions were recovered. Peptides used for ITC experiments were synthesized by Cosmo Bio (LIR(TP[T]), the sequence is DGTLIIDLPDSY) or by BEX (LIR[Sp], the sequence is LQFEIDDEMEPLYNQAKQMRYGDY). Briefly, the peptides (10 mg) were dissolved in 300 µL of distilled water and 10 µL of ammonium hydrate. The proteins and peptides except for HRV 3C protease were purified by size-exclusion chromatography with 20 mM HEPES at a pH of 6.8 and 150 mM sodium chloride using a Superdex 200 prep grade column or Superdex 75 10/300 GL column. HRV 3C protease was purified by ion-exchange chromatography with buffer A (20 mM Tris, pH 8.0) and buffer B (20 mM Tris, pH 8.0; 1 M sodium chloride) using HiTrap DEAE FF column (Cytiva, 17505501). Purified HRV 3C protease was supplemented with 20% glycerol and 1 mM dithiothreitol and stored at −80°C until use.

Crystallization and diffraction data collection
All crystallizations were performed at 20°C using the sittingdrop vapor-diffusion method by mixing protein and reservoir solutions in 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-GABARAPL2 fusion protein, 40.931 mg/ml protein was mixed with 8% polyethylene glycol 3000 (Hampton Research, HR2-604) and 0.1 M sodium citrate at pH 5.8. The crystals were soaked in cryoprotectant and frozen in liquid nitrogen. Cryoprotectants for SpHfl1 LIR-GABARAPL2 and TP53INP2 LIR-GABARAP were prepared by supplementing each reservoir solution with 25% 2-methyl-2,4-pentanediol and 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 X4 M detector attached to the beamline BL-1A at a wavelength of 1.1000 Å. Diffraction data were indexed, integrated, and scaled using energydispersive X-ray spectroscopy [42].

Structure determination
The structures of the SpHfl1 LIR-GABARAPL2 and TP53INP2 LIR-GABARAP fusion proteins were determined by the molecular replacement method using the Phenix program [43]. GABARAP (PDBID:1GNU) and GABARAPL2 (PDBID:4CO7) structures were used as the search models. Crystallographic refinement was performed using Phenix and Coot programs [43,44]. All structural images were prepared using PyMOL Molecular Graphics System v2.0. Atomic coordinates and structure factors for LIR (Sp)-GABARALL2 fusion and LIR (TP)-GABARAP fusion were deposited in the Protein Data Bank (PDB, http://www.rcsb.org/) under PDB ID codes 7YO8 and 7YO9, respectively.

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 (500 µM) were prepared as injection samples. The LC3 family and GABARAP subfamily proteins (50 µM) were prepared as cell samples. After a test injection of 0.4 µL, titration involved 18 injections of 2 µL of the 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 dilution heat. MicroCal Origin 7.0 software was used for data analysis. Thermal measurement data for the first test injection of syringe samples were removed from the analysis. Thermal titration data were fitted to a single-site binding model that was used to determine thermodynamic parameters, such as enthalpy, K d , and stoichiometry of binding (N). When the fit did not converge owing to weak interactions, N was fixed at 1.0. The error for each parameter represents the fitting error.

Quantification and statistical analysis
All data are presented as mean ± SEM and were obtained from experiments performed in triplicate or higher. The Kolmogorov-Smirnov normality test was performed to check the Gaussian distribution of the group. For two-group comparisons, a twotailed Student's t-test was used. For multiple group comparisons, 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 performed using GraphPad Prism 6 software. Statistical significance was set at P < 0.05.