Membrane binding and homodimerization of Atg16 via two distinct protein regions is essential for autophagy in yeast

Abstract

Macroautophagy is a bulk degradation mechanism in eukaryotic cells. Efficiency of an essential step of this process in yeast, Atg8 lipidation, relies on the presence of Atg16, a subunit of the Atg12–Atg5-Atg16 complex acting as the E3-like enzyme in the ubiquitination-like reaction. A current view on the functional structure of Atg16 in the yeast S. cerevisiae comes from the two crystal structures that reveal the Atg5-interacting α-helix linked via a flexible linker to an α-helix that assembles into a homodimer. This view does not explain the results of previous in vitro studies revealing Atg16-dependent deformations of membranes and liposome-binding of the Atg12–Atg5 conjugate upon addition of Atg16. Here we show that Atg16 acts as both a homodimerizing and peripheral membrane-binding polypeptide. These two characteristics are imposed by the two distinct regions that are disordered in the nascent protein. Atg16 binds to membranes in vivo via the amphipathic α-helix (amino acid residues 113–131) that has a coiled-coil-like propensity and a strong hydrophobic face for insertion into the membrane. The other protein region (residues 64–99) possesses a coiled-coil propensity, but not amphipathicity, and is dispensable for membrane anchoring of Atg16. This region acts as a Leu-zipper essential for formation of the Atg16 homodimer. Mutagenic disruption in either of these two distinct domains renders Atg16 proteins that, in contrast to wild type, completely fail to rescue the autophagy-defective phenotype of atg16Δ cells. Together, the results of this study yield a model for the molecular mechanism of Atg16 function in macroautophagy.

Graphical Abstract

Introduction

Macroautophagy, hereafter autophagy, is an evolutionarily conserved, nutrient-recycling pathway that targets superfluous cytoplasmic material and damaged organelles for degradation [1]. Autophagy is initiated by nucleation of Atg9 and COPII vesicles and protein assemblies, such as the Atg1 and Vps34 complexes, which produces a cup shaped structure, the phagophore [2–4]. A repertoire of proteins from the autophagy machinery is then recruited to expand the phagophore. The growing phagophore ultimately engulfs the cytoplasmic cargo in the transient, double-membrane autophagosome [5]. An essential step for expansion of the phagophore membrane is conjugation of Atg8 (LC3 or GABARAP in more complex eukaryotes) to the primary amino head group of phosphatidylethanolamine (PE). This ubiquitination-like reaction requires an E1-E2-E3-like enzymatic cascade, where Atg7, Atg3, and the Atg12–Atg5-Atg16 complex act as the E1, E2, and E3 ligase, respectively [6, 7].

Next to these ligases, efficient Atg8 lipidation in vivo requires an additional protein, Atg21 [8, 9]. This β-PROPPIN interacts with Asp101 and Glu102 of Atg16 and recruits the protein to PtdIns3P-enriched membranes [9]. Thus, Atg5 and Atg21 are the sole proteins currently known to directly interact with Atg16 in the process of Atg8 lipidation, when the Atg12–Atg5-Atg16 complex localizes to the surface of the phagophore membrane, but not to the autophagosome [10].

Atg8 can be conjugated to PE on small unilamellar vesicles (SUVs) in vitro, and the minimal protein set required for the efficient reconstitution reaction includes Atg3, Atg7, Atg8, and the Atg12–Atg5-Atg16 complex [11]. A previous study showed that Atg16 facilitates Atg8 lipidation in vitro and that this protein significantly promotes binding of the Atg12–Atg5 conjugate to liposomes [12]. It has also been reported that giant liposomes exhibit Atg16-dependent deformations of the membrane [13], indicating that Atg16 may contain an amphipathic α-helix and bind to membranes directly. In this study, we show that Atg16 is in fact a peripheral membrane-binding protein. We reveal how this membrane-binding characteristic of Atg16 is structurally executed and how it is manifested, along with other protein features, in autophagy. Finally, we propose a model for the molecular mechanism of Atg16 function in autophagy. This model explains the previously observed effects of Atg16 on liposomes, a behavior that the crystal structure cannot elucidate. Finding that nascent Atg16 is an intrinsically disordered protein clarifies the infeasibility of crystallization of the full-length protein from S. cerevisiae. We also reveal how the amino acid sequence of Atg16 dictates functions of the disordered regions that transition into the coiled-coil domain and membrane-inducible amphipathic helix. The study presented here provides critical information that will enable us to visualize Atg16 in a complex with Atg5 on the phagophore. This depiction advances our understanding of the Atg8 lipidation process, and adds information that, so far, has not been possible using the knowledge obtained from previous biochemical or crystallographic studies.

Results

Atg16 is a peripheral membrane-binding protein

In the Atg12–Atg5-Atg16 complex from yeast (Fig. 1A), Atg5 is a ubiquitin-like polypeptide that is conjugated to Atg12 and that functions as an essential protein in Atg8 lipidation. Liposome co-sedimentation assays with recombinant Atg5 show that electrostatic forces mediate the interaction of Atg5 with lipids [12]. This assay is consistent with the behavior of Atg5 in vivo, wher the protein associates primarily with the membrane fraction, as evidenced by a previous study [14], and by the subcellular fractionation experiment presented here; the latter was carried out using multiple-knockout (MKO) cells [15], which lack most of the ATG genes including those involved in the Atg8 and Atg12 conjugation systems, and which expressed the Atg5-MYC fusion protein (Fig. 1B, Table S1). Previous biochemical studies [14, 16] as well as the crystal structure (PDB ID: 3W1S; Fig. 1A) [17] show that Atg5 interacts directly with Atg16. This outcome is consistent with the result of our affinity-isolation experiment showing that Atg5 and Atg16 interact in MKO cells (Fig. 1C). Subcellular fractionation of MKO cells expressing Atg5-MYC and His₆-Atg16 reveals that His₆-Atg16 pellets to the membrane fraction, as does the majority of Atg5 (Fig. 1D).

Figure 1.

Atg16 from the budding yeast S. cerevisiae binds to membranes. A. A model of the Atg12–Atg5-Atg16 complex in yeast assembled using the two crystal structures, PDB IDs: 3W1S and 3A7P. B. Subcellular fractionation experiment showing that Atg5 associates with membranes in MKO cells. C. Affinity-isolation experiment confirming that Atg5-MYC and His₆-Atg16 interact in MKO cells. D. Subcellular fractionation reveals that Atg5-MYC and His₆-Atg16 associate with membranes in MKO cells. E. A domain representation of Atg16 from the budding yeast S. cerevisiae based on the crystal structures in A. F. Subcellular fractionation experiment shows that full-length Atg16 associates with the membrane fraction in MKO cells. G. A schematic model of the His₆-tagged Atg16[50–150] fragment depicted using the crystal structure (PDB ID: 3A7P). H. The Atg16[50–150] region pellets with the MKO membrane fraction in the subcellular fractionation experiment. T, total fraction; M, membrane-associated fraction; C, cytosolic fraction. Pgk1 and Dpm1 were used as controls representing a cytosolic and membrane-associated protein, respectively.

We found that His₆-Atg16 fractionates exclusively with membrane proteins, whereas a subpopulation of Atg5 remains cytosolic. Despite this different fractionation pattern, one might argue that Atg5, as a peripheral membrane-binding protein, pulls Atg16 to the membrane fraction, owing to their interaction via a single helix at the Atg16 N terminus (residues 22–46; Fig. 1E, Atg5-binding helix). To test the membrane-binding ability of Atg16, we repeated the subcellular fractionation experiment with MKO cells expressing His₆-Atg16 in the absence of Atg5. We found that His₆-Atg16 associated exclusively with membranes in MKO cells lacking, among other proteins, Atg5 and Atg21 (Fig. 1F), the only polypeptides known to interact directly with Atg16. We hypothesized that the membrane-binding ability of Atg16 may originate in the α-helical domain (amino acid residues 55–142; Fig. 1E, coiled-coil domain) that has been visualized in the crystal structure (PDB ID: 3A7P). To test this assumption, we placed a short His₆ tag on the N terminus of the Atg16[50–150] fragment (Fig. 1G); this small epitope tag should have a minimal effect on the behavior of Atg16. The subcellular fractionation experiment with this fragment in MKO cells revealed that, indeed, the 50–150 region of Atg16 exclusively associated with the membrane fraction (Fig. 1H).

Atg16 binds to membranes via a C-terminal amphipathic helix

Intermolecular interactions in parallel coil-coiled dimers are well-known to be mediated by nonpolar amino acid residues forming a hydrophobic interacting face. These residues are in the positions “a” and “d” in a schematic representation of a coiled-coil repeated pattern, known as a heptad repeat [18]. Amphipathic helices of monomeric peptides interacting with membranes possess a similar set of hydrophobic residues that, in this case, form a hydrophobic face of an α-helix that inserts into a lipid bilayer. This similarity makes it sometimes difficult to distinguish a coiled-coil type of helix from an amphipathic type, as amphipathic helices can exhibit a heptad-repeat-like pattern [19–21]. One way to recognize that a coiled-coil-like helix of a peptide is in fact amphipathic is to position the sequence of a monomeric peptide on a helical-wheel representation, where a diagram has typically four residues per one turn.

In Atg16 from S. cerevisiae, I104, I108, and V112 are exposed to the outer surface of the dimer and, therefore, cannot be involved in formation of the Atg16 coiled-coil [22]. These residues are also on the charged hydrophilic face in a helical-wheel representation of the Atg16 monomer, which excludes their involvement in membrane binding. Hydrophobic patches of amino acid residues on protein surfaces are typically not exposed to the hydrophilic cytosol; these residues tend to be hidden in hydrophobic interfaces. The immediate proximity of the hydrophobic cluster, created by I104, I108, and V112 to Asp101 and Glu102, the two Atg21-biding residues in Atg16 [9], suggests that they may mediate a hydrophobic contact with Atg21. Indeed, a recent study showed that I104, I108, and V112 mediate the Atg21-Atg16 interaction [23]. Structurally, the 100–112 segment in Atg16 separates the 55–142 domain that has been crystalized (PDB ID: 3A7P) into two regions with a heptad-repeat-like pattern, the 64–99 region and the 113–131 region. We found that the 113–131 peptide in Atg16 forms the amphipathic helix with multiple conserved residues when viewed as a monomer (Fig. 2A–C, Fig. S1); the latter suggests that this fragment may be involved in membrane binding.

Figure 2.

Membrane binding of the 113–131 region in Atg16 from yeast. A. Crystallographic (PDB ID: 3A7P) visualization of amino acid residues in the Atg16[113–131] segment subjected to site-directed mutagenesis. B. Amino acid sequence alignment for the homologous region (112–133 in S. cerevisiae) in Atg16 from various species. C. A helical-wheel representation shows that the 113–131 sequence in the monomer can fold into an amphipathic helix. Amino acid residues are depicted in a color-coded manner: yellow, hydrophobic; green, polar; blue, cationic; red, anionic; gray, polar or nonpolar depending on the pH. D. The subcellular fractionation experiment for His₆-Atg16 in MKO cells. E-G. Results of the subcellular fractionation experiments obtained with the L127A W131A, L113A L117A, and L113A L117A W131A mutants of Atg16. H. Quantification of the data from D-G. Averages and standard deviations from three independent experiments are shown. Statistical significance was tested using the unpaired two-tailed Student’s t test: ** P<0.01; *** P< 0.005. For other details, please refer to the legend to Fig. 1.

To test the lipid-binding ability of this Atg16 segment, we constructed two double mutants, L113A L117A and L127A W131A, and a triple mutant, L113A L117A W131A (Fig. 2A), where Leu or Trp residues were replaced with Ala, a hydrophobic residue lacking a side-chain necessary for insertion into a lipid bilayer (Fig. 2A). The W131 residue is the last position of the predicted amphipathic helix (Fig. 2C), and aromatic residues are found in many such helices; thus, we predict that W131 may be involved in membrane binding. Subcellular fractionation of MKO cells expressing either wild type or the mutants revealed that the residues near the flexible C terminus of Atg16 can accommodate the L127A W131A mutation, which yields the wild-type level of membrane binding and protein stability (Fig. 2D,E). In contrast to Atg16^L127A,W131A, the L113A L117A and L113A L117A W131A mutants were significantly defective in association with the membrane fraction (Fig. 2D–H) and profoundly less detectable by western blot, suggesting a substantial decrease in protein stability in cell lysates. However, these defects were not observed with a control mutant, where Leu in position 113 and 117 was replaced with Ile, another hydrophobic residue that inserts the side-chain into membrane bilayers (Fig. S2), suggesting that the specific side-chain supporting membrane insertion, not mere hydrophobicity, is essential for cellular membrane localization of Atg16. Consistent with these data, L113, L117, and W131 are positioned in the middle of the hydrophobic face in the amphipathic helix (Fig. 2C); this location is more likely to affect the nature of the hydrophobic face of the helix. To confirm that hydrophobic interactions are a major driving force in binding of Atg16 to membranes, we increased the ionic strength of the buffer, up to 500 mM NaCl. We found that a high salt concentration had no effect on membrane localization of Atg16 (Fig. S3), in support of the data in Figure 2. The high-salt conditions partially enriched the soluble, cytosolic population of L113A L117A, indicating that the unstable mutant did not form pelletable, insoluble misfolded protein aggregates, but rather was degraded in a soluble form due to insufficient anchoring to membranes. This conclusion is in agreement with a previous finding that the purified recombinant Atg16 requires 500 mM NaCl for stability in solution [22].

To confirm by a cell-free experiment that Atg16 indeed has an intrinsic membrane-binding capability, we purified recombinant Atg16[50–150] that forms a dimer, as visualized in the crystal structure (PDB ID: 3A7P). The amino acid composition of this peptide yielded poor absorption at 280 nm, which prevented us from confirming the dimeric form by analytical ultracentrifugation. Nevertheless, as expected, the Atg16[50–150] peptide failed to bind liposomes (Fig. 3A,B). Because only monomeric, not dimeric, Atg16[50–150] has the hydrophobic face of the 113–131 helix available for insertion into the lipid bilayer, we attempted to separate the dimer by adding a GFP-tag at the N terminus of the Atg16[50–150] peptide. We reasoned that the GFP tag would promote separation of the dimer into monomers and thereby allow for liposome binding of GFP-Atg16[50–150]. Indeed, we observed significant binding of the fusion protein, but not free GFP, to liposomes (Fig. 3C,D). The significantly reduced binding of the triple mutant GFP-Atg16[50–150]^{L113A,L117A,W131A} (Fig. 3E,F) to liposomes suggests that the 113–131 helix is involved in the membrane-binding mechanism. Despite a weaker signal for the wild-type, the analytical ultracentrifugation showed that the predominant forms of GFP-Atg16[50–150] and GFP-Atg16[50–150]^{L113A,L117A,W131A} were present as monomers (Fig. 3G). This finding suggests that monomeric rather than dimeric Atg16 can bind to lipids. The experiments with purified recombinant proteins (Fig. 3) are in agreement with the results obtained in cells (Fig. 1 and 2). Together, these findings show that Atg16 is de facto a peripheral membrane-binding protein that binds to membranes in its monomeric form via an amphipathic helix at the C terminus.

Figure 3.

Atg16[50–150] monomer binds to liposomes. A. Liposome sedimentation assays were conducted with Atg16[50–150] and YPL liposomes of the indicated sizes. Representative SDS-PAGE gels containing the supernatant (S) and pellet (P) fractions are shown. B. The percent of protein in the pellet fraction in A was quantified by densitometry. Error bars represent the standard deviation from three experiments. C. Liposome sedimentation assays were conducted with GFP or GFP-Atg16[50–150] and YPL liposomes of the indicated sizes. Representative SDS-PAGE gels containing the supernatant (S) and pellet (P) fractions are shown. D. The percent of protein in the pellet fraction in C was quantified by densitometry. Error bars represent the standard deviation from three experiments. E. Liposome sedimentation assays were conducted with GFP-Atg16[50–150] or the GFP-Atg16[50–150] L113A L117A W131A mutant, and YPL liposomes of the indicated sizes. Representative SDS-PAGE gels containing the supernatant (S) and pellet (P) fractions are shown. F. The percent of protein in the pellet fraction from E was quantified by densitometry. Error bars represent the standard deviation from four experiments. Statistical significance was determined by ordinary one-way ANOVA with Tukey’s multiple comparison test for B and two-way ANOVA with Sidak’s multiple comparison test in D and F. ns, not significant ****, p<0.0001. G. GFP destabilizes the dimer interface of Atg16[50–150]. Overlay of the c(s) plots from sedimentation velocity analytical ultracentrifugation for GFP-Atg16[50–150] (16 μM) and GFP-Atg16[50–150] L113A L117A W131A (20 μM). GFP-Atg16[50–150] exhibited a single peak at 4.2 S, corresponding to an average molecular mass of 56.1 kDa. GFP-Atg16[50–150] L113A L117A W131A had peaks at 4.2 and 6.8 S, corresponding to an average molecular mass of 56.6 kDa and 115.0 kDa, respectively. The expected molecular mass for the GFP-Atg16[50–150] monomer is 41.6 kDa and the expected molecular mass for the dimer is 83.2 kDa.

The 64–99 segment of Atg16 is not required for strong anchoring on the membrane

Given a good agreement of subcellular fractionation of MKO cells with liposome-binding assays (Fig. 1–3), we used the former approach to probe whether the α-helical segment at residues 64–99 in Atg16 (Fig. 4A) is needed for membrane binding of the protein. We positioned each half of the sequence on separate heptad-repeat patterns and monomeric helical-wheel diagrams (Fig. 4B,C, Fig. S4). The monomeric α-helices of the 64–81 and 82–99 regions did not yield helical amphipathicity, suggesting that these segments are not involved in membrane binding. To test this assumption, we constructed a full-length double mutant, L71A L75A, and a triple mutant, L85A I89A L99A. In addition, we replaced two highly conserved cationic residues (Fig. S5A,B), K94 and R98, with Ala to exclude an involvement of electrostatic interactions between the positive surface of Atg16 and negatively-charged phospholipids (Fig. 4A–C). Subcellular fractionation of MKO cells expressing the Atg16 mutants showed that the three mutants localized to the membrane fraction similar to the wild type (Fig. 4D–G), suggesting that neither the 64–99 segment nor the cationic patch in Atg16 are required for the stable anchoring of Atg16 to membranes.

Figure 4.

The 64–99 region in Atg16 from yeast is not required for binding to membranes. A. Crystallographic (PDB ID: 3A7P) visualization of amino acid residues in the 64–99 segment that were subjected to site-directed mutagenesis. B. A heptad-repeat representation of the 82–99 helical segment in the Atg16 dimer. C. A helical-wheel representation of the monomeric 82–99 sequence. Amino acid residues are depicted in a color-coded manner: yellow, hydrophobic; green, polar; blue, cationic; red, anionic. D-F. Results of subcellular fractionation obtained with the K94A R98A, L71A L75A, and L85A I89A L99A mutants of Atg16. G. Quantification of the data from D-F. Averages and standard deviations from three independent experiments are shown. Statistical significance was tested using the unpaired two-tailed Student’s t test: ** P<0.01; *** P< 0.005. For other details, please refer to the legend to Fig. 1.

Dimerization of Atg16 in vivo

Visualization of the 55–142 domain in the Atg16 crystal structure (Fig. 1G, PDB ID: 3A7P) suggests that purified, recombinant Atg16 forms a parallel coiled-coil dimer. Given our finding that Atg16 utilizes the 113–131 region for localization to cellular membranes, we asked whether dimerization of Atg16 is a physiological process occurring in vivo. For this purpose, we produced a mutant (L71C) in the 64–99 segment that carries Cys in place of Leu in the “d” position of the heptad repeat (Fig. S4). The “d” position is the most optimal location for the placement of cysteine that in the oxidized state engages in a disulfide bridge and, thereby, stabilizes coiled-coil dimers [24] without the need for running a native gel. We tested in vivo dimerization in MKO cells expressing His₆-Atg16^L71C under normal growth conditions and under oxidative conditions in the presence of 200 μM CuSO₄ and a high aeration of the medium. We observed a formation of dimeric Atg16 under both conditions, albeit oxidative conditions improved the efficiency of in vivo dimerization (Fig. 5A). These data show that homodimerization of Atg16 is in fact a physiological process taking place in cells.

Figure 5.

Atg16 dimerizes in vivo. A. The subcellular fractionation experiment with the Atg16 L71C mutant. The Leu71 to Cys replacement introduces a disulfide bridge at a “d” position in the coiled-coil and, thereby, stabilizes the Atg16 dimer in vivo under normal conditions in regular selective synthetic medium (left) and under oxidative conditions (right), which promote oxidation of cysteine. The oxidative conditions were induced by 200 μM CuSO₄ and a high aeration of medium. The gel was run in the absence of reducing agents. For other details, please refer to the “Materials and Methods” section. B. Affinity-isolation experiment with MKO cells expressing MYC-Atg16 from the pCuMYC-Atg16 (426) plasmid and His₆-Atg16 from the pCuHis₆-Atg16 (424) plasmid, where the latter was the wild type or L85A I89A L99A variant. A scrambled sequence of 14 amino acids fused to the His₆ tag on the pRS424 plasmid with the CUP1 promoter was used as an empty vector. For other details, please refer to the legend to Fig. 1.

The result in Figure 5A also raised a possibility that the 64–99 segment in Atg16, which is not required for membrane binding (Fig. 4), may be needed for dimerization of Atg16 in vivo. To probe this hypothesis, we expressed MYC-Atg16 in MKO cells along with the wild-type or mutant (L85A I89A L99A) variant of His₆-Atg16. The mutant variant results in the replacement to Ala in the positions “a” and “d” of the heptad repeat (see Fig. 4B). Taking into consideration the mechanism proposed for coiled-coil folding [18, 25, 26], this triple mutant presumably disrupts the beginning of the Leu-Ile-mediated zippering into the coiled-coil structure. We carried out an affinity-isolation experiment where we probed the presence of MYC-Atg16 after affinity isolation of His₆-Atg16 using Ni-NTA agarose; His₆-tagged to a nonspecific sequence of 14 amino acids was used as a negative control. The affinity-isolation experiment showed that His₆-Atg16 interacted with MYC-Atg16 (Fig. 5B), whereas the Atg16^{L85A,I89A,L99A} mutant was extremely inefficient in pulling down MYC-Atg16, as was the nonspecific sequence in the negative control. The inefficient interaction between MYC-Atg16 and His₆-Atg16^{L85A,I89A,L99A}, or the two His₆-Atg16^{L85A,I89A,L99A} mutants, is also apparent in the input, where the protein level of the His-tagged mutant is lower than that of the His-tagged or MYC-tagged wild type. This difference in protein levels is understandable, because the mutant, unable to fold into a stable coiled-coil, is more prone to degradation. Together, the results in Figure 5 suggest that Atg16 homodimerizes in vivo and this process requires the zippering mechanism that involves L85, I89 and L99.

Membrane binding and dimerization are essential for Atg16 function in autophagy

After we established the two features that Atg16 exhibits in vivo, that is, membrane binding and homodimerization (Fig. 2–5), we tested whether these characteristics are important for nonselective autophagy. We used a cytosolic zymogen of the Pho8 phosphatase that lacks the first 60 amino acids of its N terminus (Pho8Δ60) and that relies on nonselective autophagy for vacuolar localization, where the zymogen is activated [27]. The Pho8Δ60 activity assays showed a substantial increase in the activity for WT and all mutated variants of Atg16, except for those Atg16 mutants that exhibited a defect in homodimerization or membrane binding (L85A I89A L99A, L113A L117A, or L113A L117A W131A) (Fig. 6A).

Figure 6.

Assays for autophagy activity of the Atg16 mutants. A. Assay assessing Pho8Δ60 activity for wild-type and mutated His₆-Atg16 expressed on the pRS424 plasmid under the control of the CUP1 promoter in yeast cells (HPY088). B,C. GFP processing assay for the same mutants and cells as in A. D,E. Atg8–PE conjugation assay carried out in atg16Δ cells (HPY038) expressing wild-type and various mutants of Atg16. One representative western blot, in B and D, reflecting the data used for quantification is shown. Averages and standard deviations from three independent experiments are presented. Statistical significance was tested using the unpaired two-tailed Student’s t test: ** P<0.01; *** P< 0.005. Averages are normalized to that of WT (100%).

To test functionality of these latter three mutants, relative to the other Atg16 variants, by an alternative approach, we performed the GFP-Atg8 processing assay that probes for the appearance of free GFP after nitrogen starvation. GFP-Atg8 is a marker for the autophagosome because a portion of the (GFP-)Atg8 population is retained within the completed vesicle. The GFP moiety has a higher stability than Atg8 in the vacuole lumen after its fusion with the autophagosome, and thus can be used to monitor autophagic flux [28]. A strong band corresponding to free GFP was observed by western blot for wild type and for the K94A R98A or L71A L75A mutant of Atg16 (Fig. 6B,C). Cells expressing the L127A W131A mutant exhibited a slight decrease in GFP-Atg8 processing. In contrast, expression of His₆-Atg16^{L85A,I89A,L99A}, His₆-Atg16^L113A,L117A, or His₆-Atg16^{L113A,L117A,W131A} rendered yeast cells completely defective in autophagy flux, as no free GFP was detected by western blot (Fig. 6B,C). It is noteworthy that the L113I L117I mutant that, in contrast to L113A L117A, associated with cellular membranes (Fig. S2) exhibited a wild-type level of autophagy flux (Fig. S6A).

Atg16 interacts with the Atg12–Atg5 conjugate and together they act as the E3 enzyme in conjugation of Atg8 to PE. Therefore, we asked whether a defect in the autophagy flux observed for L85A I89A L99A, L113A L117A, or L113A L117A W131A (Fig. 6A–C) is due to a defect in Atg8 lipidation. Analysis of Atg8–PE by western blot showed that these three mutants of Atg16 were unable to promote Atg8 lipidation, in contrast to wild type and the other Atg16 variants (Fig. 6D,E). This finding suggests that the efficient Atg8–PE conjugation reaction and, consequently, autophagy, requires both dimerization and membrane association of Atg16.

Atg16 is an intrinsically disordered polypeptide

A low cellular stability of His₆-Atg16^L113A,L117A and His₆-Atg16^{L113A,L117A,W131A}, the mutants that failed to associate with membranes (Fig. 2), suggests that anchoring of Atg16 on the membrane ensures the stability of the protein. To test this hypothesis, we constructed a truncation mutant His₆-Atg16[1–106] that lacks the C-terminal membrane-binding amphipathic helix, but retains the 64–99 region that could, in theory, “lock” the mutant into a stable dimer, unless the stable membrane anchoring is a prerequisite for initiation of the Leu-zippering mechanism. The subcellular fractionation experiments showed that only a trace amount of the His₆-Atg16[1–106] fragment was detectable in the membrane-associated fraction by western blot (Fig. 7A,B), suggesting that the peptide was mostly degraded in a similar fashion as L113A L117A or L113A L117A W131A (Fig. 2). Moreover, the Atg16[1–106] fragment could not be stabilized even when expressed in MKO cells in the presence of Atg5 (Fig. 7A,B). This finding reveals that stable anchoring of Atg16 on the membrane and, thus, protein stability, is Atg5 independent.

Figure 7.

Atg16 is an intrinsically disordered polypeptide that folds into a dimer bound to cellular membranes. A. Atg16[1–106] lacking the amphipathic helix, but not the dimerization domain, is an unstable Atg16 fragment. B. The Atg16[1–106] segment cannot be stabilized or efficiently rescued on membranes by Atg5. Subcellular fractionation was carried out as described in the legends to Fig. 1–3. Quantification of data for this mutant was not feasible due to a very low protein level detected by western blot. C. Probing stability of His₆-Atg16 and PA-Snx4/Atg24 solubilized from membranes with 0.5% Triton X-100 and incubated in the MKO cell lysate for 3 h at 4°C. The cytosolic protein Pgk1 was used as a loading control. Atg16 is visualized using the crystal structure (PDB ID: 3A7P), whereas Snx4 is shown as the Phyre2 model. D. The subcellular fractionation of the L71C L120C and L71C L85C L99C L120C mutants of Atg16. Disulfide bridges, at a “d” position in the in the heptad-repeat pattern, allow for Atg16 dimerization in vivo under oxidative conditions without interfering with membrane binding. For other details, please refer to the legend to Figure 5A.

The above results raise a question of why membrane-free Atg16 is unstable. This behavior appears to be contradictory to the stable helical elements visualized in the crystal structure of the recombinant Atg16 protein (Fig. 1A). Because protein instability is associated with the absence of stable folding, we analyzed the amino acid sequence of Atg16 using several bioinformatics algorithms designed to predict protein intrinsic disorder [29–31]. The predictors showed that the amino acid composition of Atg16 yielded a propensity to protein intrinsic disorder. (Fig. S6B). It has been long established that interactions in coiled-coils generate relatively stable, soluble structures due to a long hydrophobic interface [18, 32, 33]. Yet, recent studies focused on protein folding show that these structures originate in fact in a disordered conformation. In principle, coiled-coil helices are formed via a disorder-to-order transition of disordered monomeric proteins or protein regions during their assembly in a complex [18, 34–36]. Dissociation from a destination complex reverts coiled-coil elements to an unfolded monomeric state and, thus, returns the polypeptides to a state that is highly susceptible to proteases, which is a characteristic of intrinsically disordered proteins [37].

To determine whether, in agreement with the prediction (Fig. S6B), Atg16 behaves as an intrinsically disordered polypeptide when dissociated from membranes, we carried out a stability experiment. We solubilized His₆-Atg16 from cell membranes using 0.5% Triton X-100 in buffer containing protease inhibitors. Specifically, the protease inhibitor cocktail cOmplete-EDTA free (Roche) inhibits a broad range of cysteine and serine proteases, whereas metallo- and aspartyl proteases are not inhibited. For example, proteinase A is an aspartyl protease, encoded by the PEP4 gene that functions in autophagy and targets a wide range of cargo. We assumed that a detergent treatment releases the protein from the membrane and, at the same time, exposes it to uninhibited proteases in the lysate. To assess degradation of the protein by these proteases, we split the detergent-solubilized lysate into two aliquots. The first aliquot was TCA precipitated at the beginning of the experiment, whereas the second aliquot was incubated for 3 h at 4°C and then TCA precipitated. Analysis of the protein level by western blot showed that His₆-Atg16 is very susceptible to proteolytic degradation, as no protein was detected in the 3-h treatment aliquot (Fig. 7C-left), in contrast to the 0-h treatment aliquot. As a control, we used protein A (PA)-Snx4/Atg24, a PX-BAR protein that folds into a stable structure mediated by BAR-type, but not coiled-coil-type, helices and that has both cytosolic and membrane-bound populations in cells (Fig. S7A). We found that, in contrast to His₆-Atg16, detergent-solubilized PA-Snx4/Atg24 was fully stable in the MKO lysates under the same buffer and incubation conditions that were applied to His₆-Atg16 (Fig. 7C-right). The susceptibility of membrane-free Atg16 to proteolytic degradation is in agreement with the protein sequence analysis, which suggested that Atg16 is an intrinsically disordered pre-molten globule (Fig. S6B, [38]).

The intrinsically disordered nature of nascent Atg16 and the membrane-binding ability of the 113–131 region appear to be in contrast to a stable homodimer seen in the crystal structure (PDB ID: 3A7P). To further confirm that, in cells, membrane binding upon folding, not dimerization, is the function executed by the amphipathic helix spanning residues 113–131, we used the L71C mutant, and additionally placed a second cysteine, instead of Leu120, within the helix. Cys120, at the “d” position if present in the heptad repeat, should lock the hydrophobic face of the helix by generating a second disulfide bridge and, thereby, prevent the protein from binding to the membrane; provided that in the cell the 113–131 helix folds into a stable and membrane-free coiled-coil dimer, as appears to be the case in the crystal structure. In this scenario, the L71C L120C double mutant should exhibit a strong signal associated with the soluble, cytosolic fraction on western blot, because coiled-coils, once folded, are stable structures. In the case of L71C L120C, this stable structure would be further reinforced by the two disulfide bridges. However, if Atg16 is a disordered protein that utilizes folding of the amphipathic helix upon stable binding to membranes, the L120C mutation should not interfere with this cellular localization; in this case, the L71C L120C mutant should be strongly associated with the membrane fraction. Under oxidative conditions in the presence of 200 μM CuSO₄ and a high aeration of the medium, we observed a strong western blot signal for dimeric L71C L120C in the membrane fraction (Fig. 7D). A major dimeric and negligible monomeric population of Atg16^L71C,L120C molecules on membranes indicates that the covalent linkage occurs between high-plasticity molecules bound on fluid lipid membrane when the dimer “zippers up”. A further attempt to “lock” and “rigidify” the protein by three additional disulfide bridges, in the “d” position (Fig. 4B and S4), in the quadruple mutant L71C L85C L99C L120C failed to produce a strong cytosolic population of the dimer (Fig. 7D), which would be detected if the 113–131 region could fold in cells into the coiled-coil helix. These findings show that a membrane-free, single-helical homodimer (residues 55–142) visualized in the crystal does not exist in a biologically-relevant environment. Atg16 in the cell is an intrinsically disordered protein that binds to the membrane via the amphipathic helix, which promotes dimerization.

Segments of coiled coils are sometimes engaged in forming higher-order oligomers, and these oligomers could, in theory, pellet along with the membrane fraction. However, this was not the case for Atg16. Gel filtration analysis of lysates from atg5Δ atg12Δ cells expressing Atg16 alone from a multi-copy 2μ plasmid showed in an earlier study that the membrane-free protein is monomeric [39]. We excluded formation of cellular Atg16 oligomers also in the present study using an experiment where we solubilized the Atg16^{L71C,85C,99C,120C} mutant from the membrane fraction by urea. If the mutant that accumulated in the pellet fraction (Fig. 7D) forms higher-order oligomers, the four urea-insensitive disulfide bridges in four different segments of the protein should presumably lock into place at least some of these oligomeric forms, not exclusively dimers, owing to Atg16 structural plasticity. This locking would lead to reappearance of the mutant in the membrane-associated pellet after a second centrifugation of the urea-solubilized membrane fraction, which was obtained after the first centrifugation. In contrast, if the appearance of the mutant in the membrane fraction was due to membrane binding, disruption of this interaction by urea would shift the mutant form of the protein to the cytosolic fraction. The result in Figure S7B shows that only the dimeric form of the mutant was translocated by urea treatment from the peripheral-membrane to cytosolic fraction, without affecting integral membrane proteins, such as Dpm1 (Fig. S7B). This result suggests that formation of pelletable higher-order oligomers of Atg16 is very unlikely.

Discussion

Lipidation of LC3/GABARAP in mammalian cells requires the ATG12–ATG5-ATG16L1 complex acting as the E3-like enzyme in the PE-conjugation reaction [7]. Human ATG16L1 is far more complex than its yeast ortholog. The human protein consists of the ATG5-binding domain followed by a long coiled-coil domain that mediates homodimerization. Downstream of this domain are the WIPI2- and RB1CC1-interacting regions located within an intrinsically disordered segment that is connected to the C-terminal WD40 repeat, which folds into a β-propeller. A recent study identified charged residues on the surface of the ATG16L1 coiled-coil domain that mediate electrostatic interaction with phosphatidylinositol 3-phosphate, a phospholipid enriched on autophagic membranes [40]. Another recent study discovered that ATG16L1 carries two membrane-binding domains [41]. One membrane-binding region, present only in the β-isoform of human ATG16L1, precedes the WD40 repeat, and is dispensable for starvation-induced autophagy. However, the N-terminal membrane-binding region, immediately downstream of the ATG5-binding domain, carries an amphipathic helix (amino acid residues 29–46) that inserts into a membrane bilayer, but is bound to ATG5 in the ATG5-ATG16L1 crystal structure (PDB ID: 4GDL) [42]. Interestingly, in a different crystal structure (PDB ID: 4TQ0) [43], the same amphipathic helix is found in the ATG16L1 homodimer or bound to ATG5. Apparently, the absence of membranes during crystallization of the recombinant proteins forces the amphipathic helix of ATG16L1 to conceal its hydrophobic face in the ATG5-ATG16L1 interface or in the crystal-induced ATG16L1 dimer. However, when membranes are present, this helix, expressed as an individual peptide in the absence of ATG5, binds in vitro to liposomes, and is essential for both selective and non-selective autophagy in mammalian cells [41]. To visualize the characteristics of the ATG16L1 protein, we created a plausible model of ATG16L1 in a complex with ATG5 (Fig. 8A). This model incorporates the recent finding on the ability of ATG16L1 to bind autophagic membranes via hydrophobic insertion of the amphipathic helix.

Figure 8.

Models of human ATG16L1 and yeast Atg16. A. A proposed model for ATG16L1 possessing the dimerizing coiled-coil region and bound to autophagic membranes as well as to ATG5 in human cells. The model was built using the crystal structure PDB ID: 4TQ0 for the ATG5-ATG16 dimer and the crystal structure PDB ID: 5NUV for WD40 of ATG16L1. The coiled-coil dimer was created as a homology model by the SWISS MODEL server using the 5TVB crystal structure as a template. The membrane-binding amphipathic helix that is required for autophagy is highlighted in red. B. A model for recruitment of Atg16 by Atg21 based on the results of this and previous studies. D101 and E102, likely along with I104, I108, and V112, in Atg16 mediate the interaction with Atg21 that recruits Atg16. Nascent monomers of Atg16 are anchored and accumulated on the membrane, which allows for dimerization. C. A proposed model for Atg16 dimerizing via the coiled-coil region and bound to autophagic membranes and to Atg5 in yeast cells. The model was built using the crystal structures 3W1S for the Atg5-Atg16 dimer and 3A7P for the membrane-bound and Leu-zippered α-helices. The amphipathic helix anchored to the membrane and required for autophagy is highlighted in red. D. Anchoring of Atg16 on membranes via the amphipathic helix determines only one spatial orientation of the protein (dark green) relative to the lipid bilayer and to Atg5. The membrane-free 55–142 region of Atg16, locked in a single-helix homodimer and floating in the cytosol, does not exist in cells. If present in the cell, it could adopt multiple spatial orientations, owing to the flexible linker. These multiple orientations (pale green) could lead to off-target interactions of Atg16 with nonspecific components of the cytosol, or to suboptimal orientation of the Atg12–Atg5-Atg16 complex.

For comparison, we also present a putative model of Atg16 from the budding yeast S. cerevisiae (Fig. 8B,C) that was built based on the results of the present study. We found (Figs. 1–7) that Atg16 is a coiled-coil-containing polypeptide that associates with cellular membranes via the C-terminal amphipathic helix, after recruitment by Atg21 [9]. Amphipathic helices are widely-utilized membrane-anchoring elements arising from structurally flexible regions in peripheral membrane-binding polypeptides, and are found in autophagy proteins, such as Atg2 [44], ATG3 [19], ATG14 [45], and Atg20 [46]. It is unlikely that the amphipathic helix in Atg16 is engaged in a protein-protein interaction with an unidentified protein, rather than inserted into a lipid bilayer, as no additional membrane-binding protein was present in the in vitro experiments where GFP-Atg16[50–150] peptide binds to liposomes (Fig. 3), or where Atg16 significantly improves poor liposome binding of the recombinant Atg12-Atg5 conjugate [12], or where Atg16 deforms giant unilamellar vesicles [13]. In addition, Atg16 from yeast functions only in the Atg12–Atg5-Atg16 complex as a part of the ubiquitin-like system of proteins including Atg3, Atg4, Atg7, Atg8, and Atg10 [11]. All these polypeptides are missing in MKO cells where His₆-Atg16 associates with membranes (Fig. 1, 2, 4, 5, and 7).

In human ATG16L1, the absence of membranes leads to a nonspecific concealment of the amphipathic α-helix at residues 29–46 within a protein-protein interface. We conclude that the absence of lipid bilayers during crystallization of yeast Atg16 causes the same effect, which is a nonspecific concealment of the 113–131 amphipathic helix into the Atg16 dimer, yielding crystallographic visualization of a sole α-helix in the dimer. This concealment can also explain why concentrated recombinant Atg16 or Atg16[50–150] fail to associate with membranes in liposome-binding assays [12], (Fig. 3A,B). The amphipathic helix of these recombinant proteins is simply “locked” in the dimer and unavailable for membrane insertion. Addition of the purified Atg12–Atg5 conjugate to recombinant Atg16 or attaching the GFP tag at the N terminus of Atg16[50–150] counteracts (solubilizes) the stable “locking”, that is, it forces the dimmer to fall apart, allows for interaction of monomeric forms with lipids, and ultimately causes liposome binding of the Atg12–Atg5-Atg16 complex or the GFP-Atg16[50–150] fusion peptide in the liposome-binding assays [12] (Fig. 3C,D, and G). Thus, the GFP tag in the fusion protein mimics the dimer-separating action of the Atg12–Atg5 conjugate upon binding to the purified recombinant Atg16.

Based on the electron density in the crystal, the coiled-coil domain of Atg16 from yeast was visualized as a single homogeneous α-helix, spanning amino acid residues 55–142 [22]. However, our analysis of the amino acid sequence by several predictors of secondary structure or coiled-coil formation suggests that there are two α-helices separated by a short linker (Fig. S8). The first predicted helix is in the region between amino acid residues 60 and 100, and overlaps with the 64–99 segment that is indispensable for dimerization of Atg16 (Fig. 5). The second predicted helical region (amino acid residues 110–140) overlaps with the amphipathic α-helix at residues 113–131 that is responsible for binding of Atg16 to the membrane (Fig. 2). However, it is important to note that the probability values in coiled-coil predictors, especially DeepCoil, need to be considered with caution because the algorithm provides a score for any helical structure, not just coiled-coil helices (Fig. S8).

We reported previously that proteins of the core autophagy machinery in yeast are enriched in intrinsically disordered regions [38]. Using PONDR-FIT, a meta-predictor for prediction of propensity for protein intrinsic disorder, we showed that, next to Atg16, the coiled-coil regions in Vps30/Atg6, Atg14, and Atg11 are among disordered segments in autophagy proteins [38]. Here we show experimentally that disruption of membrane association that prevents the coiled-coil formation, either via solubilization by a detergent (Fig. 7C) or via mutagenesis (Fig. 2, 4, 7A,B), renders Atg16 that is protease-susceptible, a characteristic of a disordered protein conformation. This conformation of Atg16 is in accord with characteristics of other coiled-coil proteins or protein regions that are categorized as intrinsically disordered proteins or protein regions that undergo a disorder-to-order transition upon folding into a coiled-coil structure [34]. The proposal that Atg16 is intrinsically disordered is also in agreement with experimental studies showing that disassembled coiled-coil regions of SNARES are in fact unstructured [35, 47, 48]. Despite all this evidence, it may be difficult to think that the Atg16 coiled-coil dimer, as seen in the crystal structure (PDB ID: 3A7P), is not stable and dissociates over time into an unstructured conformation. At this point it should be emphasized that the dimer in the crystal is a mutant, not the wild type. As the authors of the crystallographic study [22] pointed out, crystallization of full-length Atg16 is not possible, and can be achieved only by introducing the double mutation that replaces fully conserved Asp101 and Glu102 (Fig. S5) to Ala. The two hydrophobic Ala residues in place of the two anionic residues on the surface increase the overall protein compactness. Moreover, despite using the full-length Atg16^D101A,E102A protein, only residues 55–142 show a defined electron density; the entire N terminus (residues 1–54; Fig. 1E) is disordered and omitted from the model [22].

The Atg8-lipidation assays presented here along with the assays assessing the autophagy flux (Fig. 6) clearly show that the membrane anchoring, while reflecting the protein’s functional state (Fig. S9), is not the only characteristic required for Atg16 function in autophagy. The Atg16 protein also needs the ability to dimerize, in agreement with a previous study [39]. How does Leu-zippering of the 64–99 region contribute to autophagy? One possibility is that dimerization of the 64–99 region mediates tethering of vesicles along with, or before, the action of SNARE proteins operating in autophagy [49]. In any case, anchoring of Atg16 to the membrane is a prerequisite for its dimerization, as evidenced by a high cellular instability of the Atg16[1–106] fragment (Fig. 7A,B) that does possess the dimerization region, but is lacking the membrane-anchoring domain. Why does Atg16 bind to membranes first and then dimerize? This phenomenon can be explained by a molecular mechanism that has been established to take place during coiled-coil folding [18, 25, 26]. In particular, coiled-coils contain short sequences with α-helical propensity that, per se, do not dimerize. These sequences, also referred to as trigger helices, act as nucleation sites responsible for initial stabilizing interactions, and ensure a correct, in-register folding of coiled-coils. These α-helical elements, proposed to drive dimerization of the Gcn4 transcription factor and many other coiled-coil proteins, gain certain thermodynamic stability once folded, but are not fully structured in monomeric heptad-repeat-containing peptides [18, 25, 26]. Interestingly, the amino acid sequence alignment of Atg16 with Gcn4 shows that this essential α-helical element can be located in between residues 111 and 123 of Atg16 (Fig. S5C), a segment that involves Leu113 and Leu117, and that is highly homologous not only in Atg16 from fungi, but also in the orthologs from more complex eukaryotes (Fig. S5B,C). For human ATG16L1, it remains to be elucidated whether this segment drives in-register folding of the coiled-coil domain. In yeast, membrane anchoring appears to be the stabilizing mechanism for the “trigger sequence” in Atg16 that ensures correct Leu/Ile-zippering up of the coiled-coil domain. This requirement would explain why the mutation of L113 and L117 to nonpolar Ala, but not to Ile, or the deletion of the “trigger sequence” in the Atg16[1–106] fragment is so detrimental with regard to membrane binding and autophagic function (Fig. 2, 6, 7, Fig. S2, S6A).

The Atg16[1–106] fragment or the mutants carrying Ala113 and Ala117 lack the residues that can anchor the protein stably on the membrane. Therefore, a trace amount of these Atg16 variants detected by western blot in the membrane-associated fraction (Fig. 2 and 7A,B) can only be ascribed to an adventitious, dysfunctional contact with membranes. The molecular mechanism for this adventitious behavior is embedded in the structural plasticity of Atg16. The disordered nature of the protein revealed here (Fig. 7, S6B) is in agreement with the amino acid sequence analysis classifying Atg16 as a pre-molten globule [38]. It has been established that intrinsically disordered pre-molten globules fold only upon binding and do not have the funneled free energy landscape. The free energy landscape of pre-molten globules is rather flat with multiple local minima corresponding to an ensemble of conformers with residual helicity [50–52]. Considering this type of landscape for the Atg16 pre-molten globule (Fig. S10A) illustrates the difference between stable anchoring to the membrane, via folding upon insertion of the hydrophobic face of the amphipathic helix into a lipid bilayer (Fig. S10B), and an adventitious contact with the membrane. The latter applies to mutated conformers lacking the essential anchoring residues and failing to stably fold upon binding (Fig. S10C). The conformers that remain unprotected by the membrane are susceptible to proteases and degradation (Fig. S10C).

The binding of Atg16 to membranes solves a long-puzzling issue, that is a spatial orientation of the α-helical 55–142 region in Atg16 that is connected via a flexible linker to the Atg5-binding domain (Fig. 1A). Plasticity of this flexible linker can, in principle, allow for many nonspecific orientations of the membrane-free 55–142 region (depicted by pale green in the model; Fig. 8D), provided that this region existed in cells as it is visualized by the crystal structure. The membrane-anchoring of Atg16 rules out the existence of the membrane-free 55–142 domain that floats in the cytosol as a single-helical coiled-coil homodimer. The stable binding of Atg16 to the membrane determines one specific orientation of the C terminal region (dark green in Fig. 8D) and, thereby, in cells, eliminates possible off-target interactions of the Atg16 C terminus with proteins in the cytosol. Additionally, stable anchoring of Atg16 can increase the membrane-binding affinity for the entire Atg12–Atg5-Atg16 complex, and in turn ensure efficient Atg8–PE conjugation.

In summary, the results from this and previous studies [9, 16, 17, 22] reveal the plausible molecular mechanism for how Atg16 from S. cerevisiae folds and functions (Fig. 8B, C). We propose that the nascent Atg16 protein is intrinsically disordered pre-molten globule. After the recruitment by Atg21 to PtdIns3P-enriched membranes, a cascade of disorder-to-order transitions, induced by the protein-membrane and protein-protein interactions, first stably anchors Atg16 to the membrane and then “zippers up” into the dimer and allows for binding to Atg5. These processes remodel the domains with a high structural plasticity into the three α-helical elements, which are the amphipathic helix, the coiled-coil helix, and the Atg5-binding helix. Such a remodeling is essential for Atg16 function in Atg8 lipidation and, thus, autophagy.

Materials and Methods

Yeast plasmids, strains and media

pCuHis₆-Atg16 (pRS424), pCuMYC-Atg16 pRS426), and pCuAtg5-MYC (pRS426) were constructed using the FastCloning technique [53]. Other plasmids were generated from pCuHis6-Atg16 by site-directed mutagenesis [54]. Strains used in this study are listed in Table S1. Yeast cells were grown synthetic minimal medium (SMD; 0.67% yeast nitrogen base, 2% [w:vol] glucose, and auxotrophic amino acids and vitamins). Autophagy was induced by shifting the cells to nitrogen starvation medium (SD-N; 0.17% nitrogen base without ammonium sulfate or amino acids, and 2% [w:vol] glucose).

Bacterial plasmid construction

Atg16 residues 50 to 150 (Atg16[50–150]) was synthesized with an N-terminal 12 histidine, GFP tag, and an HRV-3C protease cleavage site, and cloned into pET24(+) using BamHI and HindIII by Twist Bioscience. His12-GFP-Atg16[50–150] L113A L117A W131A was produced using a Q5 site-directed mutagenesis kit (NEB, E0554S).

His-tag affinity-isolation experiments.

Cells (50 OD₆₀₀ units) were lysed in 0.5 ml of lysis buffer (1x PBS [137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 2 mM K₂HPO₄, pH 7.4], 0.2 M sorbitol, 1 mM MgCl₂, 0.5% Triton X-100, 1 mM PMSF, cOmplete EDTA-free protease inhibitor) with glass beads in 6 cycles of vortexing for 45 s with 2-min intervals of incubation on ice between each vortexing cycle. Cell debris were removed by centrifugation at 4000 × g for 5 min. An aliquot of the supernatant fraction (20%) was TCA precipitated as the Input. The remaining supernatant was incubated with 100 μL Ni-NTA agarose (Qiagen, 30210) for 3 h at 4°C. After 4 washes with ice-cold lysis buffer, the proteins were eluted by incubating the agarose for 15 min at 55°C with SDS-PAGE buffer containing 500 mM imidazole. The eluted proteins were analyzed by western blot with anti-polyhistidine monoclonal antibodies (Sigma, H1029) and with anti-MYC/c-Myc monoclonal antibody (Sigma, M4439).

Subcellular fractionation

Cells (15 OD₆₀₀ units) were lysed in 0.4 ml of PS200 buffer (20 mM PIPES-KOH, pH 6.8, 200 mM sorbitol, 5 mM MgCl₂, 1 mM PMSF, cOmplete EDTA-free protease inhibitor) or PS100 (20 mM PIPES-KOH, pH 6.8, 100 mM sorbitol, 5 mM MgCl₂, 150 mM NaCl, 1 mM PMSF, cOmplete EDTA-free protease inhibitor) with glass beads in 5 cycles of vortexing as described above. Cell debris were removed by centrifugation at 1000 × g for 10 min. An aliquot of the supernatant fraction (25%) was TCA precipitated as the Total lysate. The remaining supernatant was centrifuged at 13,000 × g for 10 min to separate the cytosolic and membrane-associated fraction. An aliquot of the supernatant fraction (30%) was TCA precipitated as the Cytosolic fraction. Both TCA precipitated samples were washed twice with acetone and air dried. The pellet (membrane-associated) fraction and the air-dried samples were resuspended in SDS-PAGE buffer containing 6 mM PSMF and incubated for15 min at 55°C before analysis by western blot. The western blot analysis of the dimeric L71C, L71C L120C, and L71C L85C L99C L120C mutants was carried out using SDS-PAGE buffer containing 6.5 M urea, 5% SDS, bromophenol blue, and 6 mM PMSF. Reducing agents (such as DTT, β-mercaptoethanol and Na₂CO₃), that break disulfide bonds and create sulfhydryl groups, were omitted. Pgk1 and Dpm1, which were used as cytosolic and membrane-associated controls, respectively, were detected on western blot using anti-Pgk1 antibody (generously provided by Dr. Jeremy Thorner, University of California, Berkeley) and anti-Dpm1 monoclonal antibody (Invitrogen, A6429).

Yeast in vivo assays

The GFP-Atg8 processing assay, Pho8Δ60 activity measurement, and Atg8-lipidation assay were performed as described previously [27, 28, 46]. Cells (1 OD₆₀₀ units) grown in selective nutrient-rich medium were shifted to nitrogen-starvation medium for 2 h (GFP-processing) or 4 h (Pho8Δ60 assay and Atg8 lipidation). Free GFP and Atg8–PE were analyzed by western blot. Densitometry quantification of western blots was performed using ImageJ software.

Bioinformatics analyses

Propensity for protein intrinsic disorder was predicted using the predictors IUPred2A, PONDR-FIT, and PrDOS [29–31]. The Atg16 secondary structure was predicted using the PREDATOR method [55], JPred server [56], or CFSSP Chou & Fasman method [57, 58]. Propensity of Atg16 for coiled-coil formation was predicted using the Multicoil [59] or DeepCoil [60] server. Protein homology models were created using the SWISS MODEL [61] or Phyre2 server [62]. Crystal structures were visualized using the PyMOL Molecular Graphic System, Version 2.0 Schrödinger, LLC.

Recombinant protein expression and purification

His12-GFP-Atg16[50–150] (referred to as GFP-Atg16[50–150] in the results section) and His12-GFP-Atg16[50–150] L113A L117A W131A (referred to as GFP-Atg16[50–150]^{L113A,L117A,W131A} in the Results section) were transformed into One Shot BL21 star cells (Invitrogen, C601003). Cultures were grown in LB Broth (Fisher Scientific, BP1426) to an optical density at 600 nm (OD₆₀₀) of approximately 0.6 at 37°C with shaking at 220 rpm. Cultures were cooled for 20 min at 4°C. Protein expression was induced by the addition of 1.0 mM isopropylthio-β-D-galactose (IPTG; IBI Scientific, IB02125). The cultures were then grown for an additional 16 h at 18°C with shaking at 220 rpm. Cells were harvested and stored at −80°C. Cell pellets were thawed and resuspended in 50 mM Tris, pH 8.0, 500 mM NaCl, 1% (v:v) Triton X-100 (VWR, AAA16046-AE), 5 mM MgCl₂, 1 mM phenylmethanesulfonyl fluoride (PMSF; VWR, 0754), and cOmplete Mini EDTA-free protease inhibitor tablets (Roche, 11836170001). Cells were lysed by passing the sample through a French Press (Thermo Electron, FA-032) four times at 4°C. Lysates were cleared by centrifugation at 40,000 × g for 50 min at 4°C. The supernatant was applied to TALON metal affinity resin (Clontech, 635504), which was preequilibrated with 50 mM Tris, pH 8.0, 500 mM NaCl. The resin was washed with 50 mM Tris, pH 8.0, 500 mM NaCl, 2.5 mM imidazole, and the protein was eluted in 50 mM Tris, pH 8.0, 500 mM NaCl, 200 mM imidazole.

To produce Atg16[50–150] without the His12-GFP tag, fractions containing protein were pooled and applied to an anti-GFP nanobody resin [63] that had been pre-equilibrated with 50 mM Tris, pH 8.0 and 500 mM NaCl. The resin was washed with 50 mM Tris, pH 8.0, 500 mM NaCl, then 20 mM Tris, pH 8.0, 100 mM NaCl. HRV-3C protease was added to the resin and incubated by rocking for two h at 4°C to cleave Atg16[50–150] from the resin. Fractions containing protein were pooled and applied to a HiLoad Superdex 75 PG column equilibrated in 20 mM Tris, pH 8.0, 100 mM NaCl, 0.2 mM tris(2-carbocyethyl)phosphine hydrochloride (TCEP; VWR, 97064–848). Fractions containing purified protein were pooled and concentrated.

To produce His12-GFP-Atg16[50–150] or His12-GFP-Atg16[50–150] L113A L117A W131A, fractions containing protein were pooled and applied directly to a HiLoad Superdex 200 PG column equilibrated in 20 mM Tris, pH 8.0, 100 mM NaCl, 0.2 mM TCEP. Fractions containing purified protein were pooled and concentrated.

His6-GFP was produced using pET His6-GFP TEV LIC (1GFP), which was a gift from Scott Gradia (Addgene, 29663). His6-GFP was transformed into CodonPlus (DE3)-RIPL competent cells (Agilent). Cultures were grown in LB Broth to an OD₆₀₀ of 0.6 at 37°C with shaking at 220 rpm. Protein expression was induced by the addition of 1.0 mM IPTG. All cultures were grown for an additional 3 h at 37°C with shaking at 220 rpm. Cells were harvested and stored at −80°C. Cell pellets were thawed and resuspended in 50 mM Tris, pH 8.5, 300 mM NaCl, 0.1% (v:v) Triton X-100, 2 mM MgCl₂, 2 mM 2-mercaptoethanol (BME), 2 mM PMSF, and cOmplete Mini EDTA-free protease inhibitor tablets. Cells were lysed by passing the sample through a French Press three times at 4°C. Lysates were cleared by centrifugation at 40,000 × g for 50 minutes at 4°C. The supernatant was applied to TALON metal affinity resin, which was preequilibrated with 50 mM Tris, pH 8.5, 300 mM NaCl, 2 mM BME. The resin was washed with 50 mM Tris, pH 8.5, 150 mM NaCl, 2.5 mM imidazole, and the protein was eluted in 50 mM Tris, pH 8.5, 150 mM NaCl, 200 mM imidazole. Fractions containing protein were pooled and applied to a HiLoad Superdex 75 PG column equilibrated in 20 mM Tris, pH 8.0, 100 mM NaCl, 0.2 mM TCEP. Fractions containing purified protein were pooled and concentrated.

Liposome sedimentation assay

Yeast polar lipids extract from S. cerevisiae (YPL; Avanti Polar Lipids, 190001) were dried under a nitrogen stream for 30 min and in a vacuum oven for 18 h. Dried liposomes were resuspended in 20 mM Tris, pH 8.0, 100 mM NaCl, 0.2 mM TCEP to a final concentration of 2.5 mg/mL. Liposomes were extruded using an Avanti Mini Extruder using a 1.0 μm Nuclepore Track-Etched membrane (VWR, 15000–610), 0.4-μm Nuclepore Track-Etched membrane (Sigma Aldrich, WHA10417104), or 0.1-μm Nuclepore Track-Etched membrane (VWR, 15000–614). Then 25 μL of 10 μM purified protein was mixed with 25 μL of 2.5 mg/mL YPL liposomes. The mixture was incubated for 60 min at 4°C to afford complex formation, followed by centrifugation at 40,000 rpm for 40 min at 4°C using a TLA45 rotor. Supernatants were removed, and equal volumes of buffer were added to resuspend the pellets. Samples were run on Invitrogen Novex NuPAGE 4 to 12% Bis-Tris gels (Invitrogen, NP0322BOX).

Densitometry was performed on both the pellet (P) and supernatant (S) fractions using Image Lab v 5.1 (Bio-Rad). The supernatant and pellet band intensity were summed together to determine the total intensity of protein in each sample. The percentage of protein bound to liposomes was determined by taking the ratio of the pellet band intensity over the total intensity. Each experiment was performed in triplicate or quadruplicate as indicated. The results were averaged and plotted with error bars representing the standard deviations of the three or four independent experiments. Statistical analysis was performed using Prism software v 8.2.1 (GraphPad). Data were compared using an ordinary one-way ANOVA with Tukey’s multiple comparison test or using a two-way ANOVA with Sidak’s multiple comparison test. ****, p<0.0001.

Dynamic Light Scattering

To determine size and homogeneity of the liposome, dynamic light scattering was performed on liposomes at 1.25 mg/mL at 20°C using DynaPro NanoStar (Wyatt Technology). Light scattering data were analyzed using Dynamics v7.1.8 (Wyatt Technology).

Analytical Ultracentrifugation

Sedimentation velocity experiments were conducted at 30,000 rpm using a Beckman Proteomelab XL-A analytical ultracentrifuge and a AN-60 rotor. The samples were monitored by absorbance at 280 nm. Both samples were in 20 mM Tris, pH 8.0, 100 mM NaCl, 0.2 mM TCEP. His12-GFP-Atg16[50–150] and His12-GFP-Atg16[50–150] L113A L117A W131A were run at 16 μM and 20 μM, respectively. The solvent density (1.0029 g·mL⁻¹), viscosity (0.010169 poise), and partial specific volumes that were used for the analyses, 0.735958 mL·g⁻¹ (His12-GFP-Atg16[50–150]) and 0.730000 mL·g⁻¹ (His12-GFP-Atg16[50–150] L113A L117A W131A), were calculated using SEDNTERP [64] (http://bitcwiki.sr.unh.edu/index.php/Main_Page). The sedimentation coefficients and apparent molecular weights were calculated from c(s) analysis using SEDFIT [65].

Supplementary Material

1

Figure S1. Helical wheel representations for the homologs of the Atg16[113–131] region (numbering in S. cerevisiae) from various species.

2

Figure S2. The Atg16^L113I,L117I mutant behaves similar to wild type. Subcellular fractionation for His₆-Atg16 wild type and the L113I L117I mutant reveals no significant difference in membrane association of the proteins. For other details, please refer to the legend to Fig. 1.

3

Figure S3. Subcellular fractionation experiments under different experimental conditions. Atg16 wild type, but not the L113A L117A mutant, exclusively associates with membranes in the presence of 150 mM and 500 mM NaCl in the buffer.

4

Figure S4. Representation of the 64–81 domain in Atg16 from S. cerevisiae in the heptad repeat of the dimeric coiled-coil (left) and on the α-helical wheel of the monomer (right).

5

Figure S5. Protein amino acid sequence alignments. A. Multiple amino acid sequence alignment for the Atg16 protein from various fungi. Leu residues (black) and cationic Arg and Lys residues (blue) that were mutated to Ala in this study are indicated by arrows. B. Homology of the putative coiled-coil domains (CCD) in Atg16 from various organisms. C. The 111–123 region in Atg16 from S. cerevisiae exhibits homology to the nucleation site in Gcn4.

6

Figure S6. Functional and bioinformatics analyses of the Atg16 variants. A. GFP-processing for the L113I L117I mutant after 3 h of nitrogen starvation. A representative western blot and the quantification of data from three independent experiments are presented. B. Bioinformatics analysis of the Atg16 (wild type) amino acid sequence. Propensity for protein intrinsic disorder, as predicted by the IUPred2A method and metapredictors PrDOS and PONDR-FIT, is shown by the majority of scores being in the range of 0.4–1.

7

Figure S7. Subcellular fractionation experiments with MKO cells expressing different autophagy proteins. A. Snx4 possesses both a cytosolic and membrane-associated fraction in MKO cells. B. The subcellular fractionation of the L71C L85C L99C L120C mutant of Atg16 yielded the C1 and M1 fractions (as in Fig. 7D), and was followed by solubilization of the M1 fraction in buffer containing 3 M urea. The subsequent second centrifugation separated the C2 and M2 fractions. For other details, please refer to the legend to Figure 5A.

8

Figure S8. Prediction of the secondary structure (PREDATOR, JPred, CFSSP, Multicoil, and DeepCoil) formation in Atg16 from the budding yeast S. cerevisiae.

9

Figure S9. Appearance of Atg16 in the membrane fraction after 2 h of nitrogen starvation of yeast cells (HPY088) suggests that association with membranes is the functional state of Atg16.

10

Figure S10. A proposed molecular mechanism for the adventitious behavior of the Atg16[1–106] fragment and the L113A L117A or L113A L117A W131A mutants. A. A schematic illustration of the multi-funnel free energy landscape for pre-molten globules, including Atg16. Local energy minima correspond to conformers with residual helicity. B. Wild type or the L113I L117I mutant folds upon binding to the membrane, which results in the functional anchoring and stabilization of the protein. C. The Atg16[1–106] fragment and the L113A L117A or L113A L117A W131A mutants lack the residues essential for stable membrane anchoring. Instead, a few conformers of these Atg16 variants contact the membrane randomly. The conformers, which are unprotected by the membrane, are degraded.

Highlights:

• Atg16 in the budding yeast S. cerevisiae is an intrinsically disordered polypeptide.

• The C terminal amphipathic helix (at amino acid residues 113–131) anchors Atg16 on membranes.

• The coiled-coil domain of Atg16 (at amino acid residues 64–99) mediates homodimerization in vivo.

• Both membrane binding and homodimerization are required for Atg16 function in autophagy.

Abbreviations:

Atg

autophagy-related

MKO

multiple-knockout

PAS

phagophore assembly site

PE

phosphatidylethanolamine

PtdIns3P

phosphatidylinositol 3-phosphate

SUV

small unilamellar vesicles

TCA

trichloroacetic acid

YPL

yeast polar lipids