Endogenous proteins of the ULK1 complex target a compartment closely associated with omegasomes

The current model suggests that, during autophagy, the activated ULK1 complex translocates to ER, on sites that are close but distinct from omegasomes, where it stimulates PtdIns3P synthesis and drives the recruitment of PtdIns3P effectors such as DFCP1 and the WIPIs (Mizushima and Komatsu, 2011). This model predicts that the ULK1 complex translocates to the ER before omegasomes are formed and remains spatially adjacent to omegasomes. In order to examine this model in a human cell line, we established the localization of endogenous proteins of the ULK1 complex in HEK293 cells stably expressing the omegasome marker GFP–DFCP1 (Axe et al., 2008). We found that in starved cells, 80% of omegasomes colocalised with punctate structures formed of ULK1, Atg13 and Atg101 (three of the four proteins of the ULK1 complex) (Fig. 1A). Ring-shaped omegasomes associated at the centre of their ring with puncta decorated with ULK1, Atg13 and Atg101 (arrows in Fig. 1A), whereas non-ring-shaped omegasomes overlapped partially (arrowheads in Fig. 1A) with the ULK1 complex. A population of smaller diameter ULK1, Atg13 and Atg101 puncta did not associate with omegasomes, but we note that many such puncta were also present in non-starved cells (see control ULK1 staining in Fig. 1A). We concluded that, during starvation, the ULK1 complex translocates to puncta, which are often spatially connected to omegasomes.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

The ULK1 complex translocates to PtdIns3P-sensitive puncta during autophagy. (A) HEK293 cells stably expressing GFP–DFCP1 were starved for 1 hour, fixed and stained with antibodies against ULK1, Atg13 and Atg101. Representative images are shown. Arrowheads point to small omegasomes tangential with particles from the ULK1 complex. Arrows point to ring-shaped omegasomes, which associate in the middle with particles from the ULK1 complex. (B) HEK293 cells were starved or treated with PP242 for 1 hour, in the absence or presence of wortmannin, and immunolabelled for endogenous ULK1, Atg13, Atg101 and WIPI2. Representative confocal images are shown. Puncta were counted using Imaris. Values are means ± s.d. puncta per cell, for at least 10 different fields with 15–30 cells each. Scale bars: ∼10 µm.

The translocation of the ULK1 complex to starvation-induced puncta is responsive to PtdIns3P levels

The ULK1 complex stimulates PtdIns3P synthesis, which must be sustained for the formation of omegasomes (Di Bartolomeo et al., 2010), and the puncta associated with the ULK1 complex (hereafter referred to as ULK1 puncta) also associate with omegasomes, which are PtdIns3P-rich structures. Thus we hypothesized that the translocation of the ULK1 complex to starvation-induced puncta might be responsive to PtdIns3P levels. To test this, we examined whether in the parental HEK293 cells the translocation of endogenous ULK1, Atg13 and Atg101 to puncta under autophagy-inducing conditions is disrupted by wortmannin, which is an inhibitor of the class III phosphatidylinositol 3-kinase (Fig. 1B). As a positive control we used WIPI2, a well characterised PtdIns3P effector in this pathway (Polson et al., 2010). We found that in the absence of wortmannin, induction of autophagy by either starvation or the mTOR inhibitor PP242, increased the number of WIPI2 puncta per cell by more than sixfold compared with control conditions (Fig. 1B). Addition of wortmannin reduced WIPI2 puncta to levels similar to those in the control conditions (Fig. 1B). Similarly, under normal conditions the number of ULK1 puncta per cell was 2.1±1 and induction of autophagy, either by starvation or by PP242, increased them by approximately fourfold (Fig. 1B). Inhibition of PtdIns3P synthesis decreased the number of ULK1 puncta per cell to levels similar to the control (Fig. 1B). The same wortmannin sensitivity was seen for the other two proteins of the ULK1 complex, Atg13 and Atg101 (Fig. 1B). As it was previously shown (Axe et al., 2008), addition of wortmannin almost abolished the formation of omegasomes; under those conditions the wortmannin-resistant Atg13 puncta were largely free of omegasome components (supplementary material Fig. S1A). Finally, depletion of the Vps34 complex member Beclin1 markedly reduced the number of large puncta that were formed under starvation by the endogenous Atg13 (supplementary material Fig. S1B,C). In summary, PtdIns3P synthesis affects the targeting of the ULK1 complex to starvation-induced puncta.

GFP–Atg13 localization pattern is representative of the ULK1 complex

The colocalization between omegasomes and the ULK1 complex reported above suggests a dynamic relationship between the two. To explore this further by live-cell imaging, we established a clonal cell line stably expressing GFP–Atg13. Atg13 is a well characterised member of the ULK1 complex (Chang and Neufeld, 2009; Ganley et al., 2009; Hosokawa et al., 2009; Jung et al., 2009). We favoured it over the other proteins of the complex because it is conserved from yeast to mammals (unlike Atg101), it does not have additional functions (unlike FIP200) and it does not appear to inhibit autophagy upon overexpression (unlike ULK1). During starvation, GFP–Atg13 was recruited to puncta, whose number increased with time (supplementary material Fig. S2A and Movie 1) and colocalised completely with puncta of the endogenous ULK1 and Atg101 (supplementary material Fig. S2B,C), as well as with puncta of the early autophagy markers WIPI2 and Atg16 (supplementary material Fig. S2B). Finally, the formation of GFP–Atg13 puncta was inhibited by wortmannin treatment (supplementary material Fig. S2B). Thus the localization of the exogenously expressed GFP–Atg13 followed the pattern of the endogenous ULK1 complex, hence justifying its further use for functional analysis. To ensure that downstream autophagy response was unaffected in these cells, we used western blotting to examine the conversion of LC3-I to LC3-II, a commonly used biochemical assay for autophagy (Klionsky et al., 2012). The two clones mostly used in this work (clone13 and clone35) had the same starvation response as the parental HEK293 cells (supplementary material Fig. S2D).

PtdIns3P synthesis affects the size and lifespan of Atg13 puncta

We next wanted to examine the dynamics of Atg13 translocation to puncta under starvation and correlate them with PtdIns3P synthesis by employing the PI 3-kinase inhibitor wortmannin. We found that addition of wortmannin to starved cells dramatically reduced the total amount of GFP–Atg13 puncta after just a few minutes, without completely abolishing them (Fig. 2A,B; supplementary material Movie 2). The pre-existing puncta continued their lifecycle until they collapsed, while new, smaller puncta formed at a much slower rate. The effect of wortmannin on the number of Atg13 puncta might be explained if PtdIns3P synthesis creates a positive feedback loop that reinforces the translocation of Atg13 to puncta, therefore increasing their lifespan. To test this, we quantified the lifespan of Atg13 puncta, from the moment that they emerge until they collapse, both in the absence and presence of wortmannin. We found that the lifespan of Atg13 puncta follows a near normal distribution, and the addition of wortmannin shortens their average lifespan by more than twofold (201 and 86 seconds, respectively; Fig. 2D,E). Moreover, the size of Atg13 puncta that continued to form in the presence of wortmannin and the total amount of Atg13 recruited on those structures were significantly reduced compared with those formed under normal starvation (Fig. 2C,F,G). A similar trend was also observed for endogenous Atg101 in fixed cells (Fig. 2H,I). We concluded that Atg13 puncta require sustained PtdIns3P synthesis in order to reach their maximum size and lifespan.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

PtdIns3P synthesis reinforces the translocation of Atg13 to puncta under starvation. (A) Live-cell imaging of starved HEK293 cells stably expressing GFP–Atg13 before and after the addition of wortmannin. Representative frames are shown. (B) Number of GFP–Atg13 puncta per field over time, for one field. The arrow marks the addition of wortmannin. (C) Representative images of GFP–Atg13 puncta from emergence to collapse, in the absence or presence of wortmannin. (D,E) The spread of the lifespan of GFP–Atg13 puncta in the absence (D) or presence (E) of wortmannin. Events are defined as independent Atg13 particles formed, from their emergence until their collapse. (F,G) The maximum size of Atg13 puncta was quantified in the absence (67 events from D) or presence of wortmannin (95 events from E). In brief, lines were drawn through the centre of Atg13 particles using ImageJ, the intensity of pixels along the line was plotted and the borders of Atg13 particle were defined as the pixels with fluorescence intensity above the background. Values are means ± s.d. of the diameter (F) and the ratio of maximum to background fluorescence intensity (G) of the Atg13 particles normalized to the values in the absence of wortmannin. (H,I) The maximum size of Atg101 puncta was quantified in the absence or presence of wortmannin (spots from Fig. 1B) using the Spot Detection function of Imaris software (Bitplane). Values are means ± s.d. of the diameter (H) and the maximum fluorescence intensity (I) normalized to values in the absence of wortmannin. Scale bar: ∼10 µm.

Relationship of Atg13 puncta to ER and mitochondria

The origin of the autophagosomal membrane remains an unresolved conundrum of the autophagy field, with ER, mitochondria and plasma membrane being the main candidates (Rubinsztein et al., 2012; Tooze and Yoshimori, 2010). Part of the uncertainty is probably related to the variability of the experimental settings under which autophagy is induced. A second confounding factor is the extremely dynamic nature of autophagosomes, which allows them to appear in close proximity to many membranes. Earlier work using exogenous ULK1 indicated that starvation-induced puncta of this protein are seen in association with the ER and frequently move with the ER strands (Itakura and Mizushima, 2010). Here we re-examined this question but used Atg13, and considered only newly formed puncta that were clearly distinguishable by spinning disk confocal microscopy. In addition, because wortmannin reduces the size and lifespan of Atg13 puncta without completely blocking their formation, we postulated that particles forming in the presence of this drug might represent the earliest visible intermediate of forming autophagosomes. When comparing the spatial relationship of GFP–Atg13 puncta with ER, mitochondria and lysosomes (Fig. 3B) we found that, in the absence of wortmannin, the Atg13 particles were always formed in close association with ER (Fig. 3A example 1, D), frequently but not always close to mitochondria (Fig. 3A examples 2 and 3, D) and less frequently in the vicinity of lysosomes (Fig. 3D). Importantly, the association between the emerging Atg13 particles and ER, mitochondria and lysosomes in the presence of wortmannin was not altered (Fig. 3D,E). The spatial relationship between Atg13 puncta and ER was also examined by super resolution microscopy, which showed that the Atg13 particles are well-defined structures that associate tightly with but are distinct from the strands of ER (Fig. 3F). Taking everything into account we concluded that Atg13 puncta form in tight association with ER, and that inhibition of PtdIns3P synthesis does not alter their subcellular localization. Thus PtdIns3P is not the signal that targets Atg13 to ER, and targeting to the ER under starvation conditions in the absence of PtdIns3P is not sufficient to allow Atg13 (and by extension the ULK1 complex) to fulfil its membrane-associated cycle.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Relationship of Atg13 puncta to ER, mitochondria and lysosomes. (A) HEK293 cells stably expressing GFP–Atg13 were first either transfected for 24 hours with mCherry-dgk1 (ER reporter; ER) or mRFP-LAMP1 (lysosomes reporter; not shown), or incubated with Mitotracker Red (mito) for 30 minutes, and then starved and imaged. Representative montages of forming GFP–Atg13 puncta close to ER (example 1) close to mitochondria (example 2) or away from mitochondria (example 3) are shown. (B,C) Schematic representation of quantitative analysis of the association between GFP–Atg13 puncta and ER, mitochondria or lysosomes. Lines were drawn to connect the centre of an Atg13 punctum with the closest strand of ER, a mitochondrion or a lysosome (B), the intensities of pixels along the line for both the green and the red channel were plotted, and the distance in pixels between the peaks of the two channels was measured (C). (D,E) Distance, in pixels, between Atg13 spots and ER, mitochondria or lysosomes, for the first three frames from their emergence, in the absence (D) or presence of wortmannin (E). Events are defined as independent Atg13 particles formed, from their emergence until their collapse. For example, for D-ER, we measured 35 unique Atg13 particles emerging de novo, and plotted the distance of each to the ER strands for the first three frames. The ordering of the events in the graph is arbitrary and starts from those closest to the ER to those away from the ER. (F) HEK293 cells stably expressing GFP–Atg13 were transfected with mCherry-dgk1 for 24 hours, starved and subjected to structured illumination microscopy (SIM). Three representative examples are shown. Signals from the green and red channel are inversely pseudocoloured as red and green, correspondingly, to allow better visual appreciation of the ER. Scale bars: ∼1 µm.

Dynamic association of Atg13 with expanding omegasomes

It is clear from our data that PtdIns3P synthesis enhances the number, size and lifetime of the ULK1 particles. Given the fact that omegasomes represent a PtdIns3P platform for early autophagy events, it was of special interest to determine how they engage with Atg13 puncta. We used live-cell imaging in cells stably expressing GFP–DFCP1 and mCherry–Atg13. We found that throughout starvation mCherry–Atg13 puncta colocalised with omegasomes (Fig. 4A; supplementary material Movie 3), in agreement with the localization of the endogenous Atg13 (Fig. 1). Only one Atg13 punctum associated with each omegasome, and a typical sequence is shown in Fig. 4B; supplementary material Movie 4. Once the two particles formed (Fig. 4B yellow and white arrows) they expanded together. The omegasome often developed the hallmark ring shape, whereas Atg13 remained as a punctate structure enlarging to fill the interior of the omegasome. Interestingly, the Atg13 puncta did not stay on omegasomes throughout the lifetime of the latter: in the example shown in Fig. 4B, the Atg13 punctum started to disappear after about 3 minutes, while the omegasome was still visible (blue arrows). After the Atg13 signal disappeared, the omegasome underwent a characteristic transformation whereby its membrane elongated, suggestive of a ‘budding’ step [Fig. 4B blue arrows (see also Axe et al., 2008)]. At the end of this sequence the omegasome collapsed and largely disappeared (Fig. 4B last three frames). A schematic representation of this sequence is shown in Fig. 4C: the green line represents the duration of the GFP–DFCP1 structure, the red line the equivalent for mCherry–Atg13, and the blue symbol marks the budding stage. To determine the order of recruitment of Atg13 and DFCP1 and their respective dynamics, examples from more than 10 videos were analysed as described in Fig. 4C, always choosing events clearly starting de novo. In total, we analysed 45 such events, which are depicted in supplementary material Fig. S3. The temporal relationship between Atg13 and DFCP1 puncta was slightly variable at the beginning but similar for the subsequent steps. At the beginning of the sequence the relative occurrence of the visible Atg13 and omegasome puncta was variable or hard to determine. In the majority of events, Atg13-positive puncta and omegasomes formed nearly simultaneously or within half a minute of each other, but in some events, either the Atg13 punctum or the omegasomes had a clear precedence (supplementary material Fig. S3). Once formed, the Atg13 and DFCP1 mixed membrane expanded, typically for 2 minutes, before Atg13 disappeared. Importantly, the budding step took place almost always after the Atg13 had left the omegasome (there were only few exceptions shown for the 45 events analysed in supplementary material Fig. S3), most typically within 1 minute from the disappearance of Atg13, and was followed by the diminution and the eventual collapse of the omegasome.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Atg13 colocalises with omegasomes but separates before the autophagosomal budding step. (A) Live-cell imaging of starved HEK293 cells stably expressing GFP–DFCP1 and mCherry–Atg13. Two representative frames of different time points are shown. Arrows point to Atg13 particles that colocalise with omegasomes. (B) Representative montage showing the spatial and temporal relationship between one Atg13 particle and one omegasome. Frames were captured every 10 seconds. Yellow arrows indicate the emergence of the Atg13 particle. White arrows indicate the emergence of the omegasome. Blue arrows indicate membrane budding off the omegasome ring. (C) Schematic illustration of the temporal relationship shown in B, paralleling the lifespan of an Atg13 particle with the formation, budding and collapse of the corresponding omegasome. (D) HEK293 cells stably expressing GFP–DFCP1 and mCherry–Atg13 were transfected with CFP-LC3 for 24 hours, starved and imaged live. Two representative frames corresponding to different time points from the beginning of filming are shown. Arrows point to structures that are marked with all three proteins. (E) Representative montage illustrating an autophagosome formation sequence. Frames were captured every 10 seconds. White arrowheads indicate the emergence of the Atg13 particle and the omegasome. Yellow arrowheads indicate the structures that are marked with all three proteins. White arrows indicate the structures after the exit of Atg13. Yellow arrows point to LC3-positive membrane budding off the omegasome. The drawing illustrates the spatial relationship between the omegasome and the autophagosome of the corresponding frame. Scale bars: ∼10 µm (A,D); ∼1 µm (B,E).

Autophagosomes are formed following an ordered course of intermediate steps

In a previous study, we showed that each omegasome is the precursor structure that leads to the formation of one LC3-positive autophagosome (Axe et al., 2008), whereas in this study, we have demonstrated that each Atg13 particle associates with one forming omegasome. Taking into consideration both studies, we assumed that autophagosome formation should comprise a reproducible course of events, starting with the nucleation of ULK1 complex and ending with the budding of a mature autophagosome from the omegasome. To show this, we followed, in starved HEK293 cells, the localization of mCherry–Atg13, GFP–DFCP1 and CFP–LC3, in order to capture the spatial and temporal dynamics between the ULK1 complex, omegasomes and LC3, along the course of autophagosome formation. We found that throughout starvation, Atg13 puncta associated with both DFCP1- and LC3-positive structures (Fig. 4D, arrows; supplementary material Movie 5). As predicted, one Atg13 particle associated with only one omegasome and gave rise to a single autophagosome; a typical sequence is shown in Fig. 4E; in supplementary material Movie 6, example 1. The Atg13 punctum formed and expanded in parallel with the omegasome (Fig. 4E, white arrowheads at 40 seconds), whereas LC3 started to accumulate on the membrane enclosed by the omegasome ring approximately half a minute later (Fig. 4E, yellow arrowheads at 1 minute 10 seconds). The structure marked by the three proteins expanded with them, with the omegasome always enclosing the Atg13- and LC3-positive puncta (Fig. 4E, arrowheads at 3 minutes 20 seconds for a typical snapshot). After about 5 minutes, the Atg13 particle rapidly disappeared, and only omegasome and LC3 staining remained visible (Fig. 4E, arrows at 5 minutes 50 seconds). Within 1 minute of the exit of Atg13, the LC3-containing autophagosome began to bud off from the omegasome (Fig. 4E, arrows at 7 minutes 10 seconds) and shortly thereafter the omegasome staining disappeared, leaving behind only the LC3-containing autophagosome. This sequence verifies that the disappearance of Atg13 from the common membrane precedes the budding step whereby the autophagosome exits the omegasome. In general, this complicated sequence of events, captured here for the first time, shows a remarkable consistency in its timing characteristics. Supplementary material Fig. S4 shows a compressed course for the event just described alongside a different event captured from a different part of the cell and at a different interval during starvation (supplementary material Fig. S4, example A and example B and Movie 6). The sequence of events and their timing is fairly reproducible. Based on the live-cell imaging experiments, we concluded that sustained PtdIns3P synthesis induces the expansion of the Atg13-positive nucleation site, and initiates a reproducible course of events that leads to the formation of an autophagosome through an omegasome intermediate.

A site on Atg13 is important for lipid binding and translocation to autophagic puncta

The inhibitory effect of wortmannin on the translocation of Atg13 to puncta, and the tight association of Atg13 particles with the expanding omegasomes, indicated that the localization of Atg13 to early autophagy structures might depend on some lipid affinity. To test this, we scanned the primary sequence of Atg13 for potential domains that could interact with acidic phospholipids, and identified a cluster of one arginine and three lysine residues very near the N-terminus that is conserved in higher eukaryotes (Fig. 5A). When we simultaneously mutated all four residues into alanines, the mutant showed a reduction of 50% in number of GFP–Atg13 puncta during starvation in a mixed stable clone (Fig. 5B,C), whereas the reduction was closer to 75% for single clones derived from this mixed population (Fig. 5D,E). Moreover, the (reduced) number of GFP–Atg13 puncta formed by the mutant clones in the presence of wortmannin was further reduced compared with the wild-type (wt) clones (Fig. 5F). These results suggest that these residues play an important role in the translocation of Atg13 particles to autophagic puncta. Other sites mutagenized on the basis of potential phosphorylation or interaction sites as indicated in Fig. 5A did not interfere with punctum formation and were not considered further (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Atg13 targeting to puncta is facilitated by its affinity for acidic phospholipids. (A) Atg13 primary sequence, illustrating the amino acid stretches that are necessary for binding to its partners (Jung et al., 2009; Mercer et al., 2009). One arginine and three lysines (highlighted in the sequence) were mutated to alanines. Asterisks indicate other point mutations of conserved residues of Atg13 that did not affect localization. A sequence alignment of these residues in human, mouse, zebrafish and yeast is shown at the top. (B) HEK293 cells stably expressing the mutant Atg13 were starved and their response was compared with the wt GFP–Atg13 clone. Representative images are shown. (C) The number of GFP–Atg13 puncta per cell in B for 20 different fields, containing 5–10 cells each (values are means ± s.d.). (D) Translocation of the mutant Atg13 to puncta under starvation for stable clones. (E) The number of GFP–Atg13 puncta per cell from D, for five different fields containing 15–30 cells each (values are means ± s.d.). (F) Clones of HEK293 cells stably expressing the mutant Atg13 were starved in the absence or presence of wortmannin, and their response was compared with the wt GFP–Atg13 clones. Values are means ± s.d. of the number of GFP–Atg13 puncta per cell, normalized to the response in the absence of wortmannin, for five different fields, containing 15–30 cells each. (G) Lysates from HEK293 cells stably expressing GFP–Atg13 (clone 35) were incubated with a panel of beads coated with lipids as shown, eluted with SDS-PAGE sample buffer, and the recovered proteins were assessed by western blotting using antibodies against GFP and Atg16. (H) Lysates from clones in D were incubated with beads coated with PA, PtdIns4P (PI4P) and PtdIns3P, and the recovered proteins were immunoblotted using antibodies against GFP and WIPI2. (I) Western blots from H were quantified with ImageJ, and values of recovered wt and mutant GFP–Atg13 normalized to the control are shown. (J) Wt and mutant Atg13 in lysates from H were immunoprecipitated with antibody against GFP, and blotted using antibodies against GFP, Atg101 and ULK1. Inputs are the same as for H and the gel is simply duplicated. (K) Recombinant wt and mutant His–Atg13 were purified from E. coli, and the eluates were run in an SDS-PAGE gel and stained with Coomassie Blue. BSA was run to quantify the amounts of purified proteins. (L) Equal amounts of wt and mutant His-Atg13 from K were incubated with lipid-coated beads as in H, and recovered Atg13 was immunoblotted using antibody against Atg13. Scale bars: ∼10 µm.

Using a panel of chemically synthesised lipids covalently coupled to a solid support, which we have extensively characterised and validated in our lab (Conway et al., 2010; Delon et al., 2004; Krugmann et al., 2002; Manifava et al., 2001; Ridley et al., 2001), we tested lysates from the clonal cell line expressing the wt GFP–Atg13 for binding to PtdIns3P, PtdIns4P, PtdIns(3,4)P₂, PtdIns(3,5)P₂, PtdIns(3,4,5)P₃, phosphatidic acid (PA), diacylglycerol (DAG), cardiolipin and sphingosine. Under conditions where endogenous Atg16 did not bind to any of the beads tested, Atg13 bound to PA-, PtdIns3P- and PtdIns4P-coated beads and more weakly to those coated with PtdIns(3,4,5)P₃ (Fig. 5G). Mutant Atg13 showed decreased affinity for PA by 60% and for PtdIns3P and PtdIns4P by 80% (Fig. 5H,I). To examine whether these mutations compromise the association of Atg13 with the ULK1 complex, we immunoprecipitated the exact same lysates of wt and mutant GFP–Atg13 shown in Fig. 5H (note that input for H and J are identical and simply duplicated) with antibodies to GFP, and blotted for Atg101 and ULK1. We found that both wt and mutant Atg13 bound to endogenous Atg101 similarly (Fig. 5J), and the same was true for binding to endogenous ULK1 (Fig. 5J). We note that, in our conditions, binding of Atg13 to endogenous Atg101 is almost stoichiometric, whereas binding to endogenous ULK1 is less so. Thus, the mutations of Atg13 did not alter its interaction with Atg101 or ULK1. To compare lipid binding of the mutant with the wt Atg13 in a simpler system, we prepared the two recombinant proteins from bacteria (Fig. 5K) and examined whether they bind to PA, PtdIns3P and PtdIns4P. The recombinant wt Atg13 bound to the same phospholipids as the GFP–Atg13 from HEK293 lysates (Fig. 5L) but its affinity for PA was stronger than its affinity for PtdIns3P, whereas the affinity for PtdIns4P was even weaker (Fig. 5L; note that the lipid beads for this experiment were identical to those used in the one shown in Fig. 5H). In agreement with our hypothesis, the mutations of the recombinant Atg13 reduced substantially its affinity for all three phospholipids (Fig. 5L). We concluded that Atg13 has modest affinity for three acidic phospholipids, namely PA, PtdIns3P and PtdIns4P, which may facilitate its translocation to puncta during starvation.

Because the mutations of Atg13 compromise its binding to lipids without affecting its interaction with the other proteins of the ULK1 complex, we asked whether the mutant Atg13 has a dominant-negative effect on the autophagy response. In preliminary experiments we noted that in a stable population of mixed clones, expression of the mutant GFP–Atg13 reduced the translocation of endogenous Atg101 to starvation-induced puncta, whereas the wt GFP–Atg13 did not (Fig. 6A). In a stable population of mixed clones or in clones derived from this population, the expression of mutant Atg13 reduced the translocation of endogenous ULK1, Atg101, WIPI2 and Atg16 to puncta by ∼50% (Fig. 6B,C). Moreover, stable expression of the mutant GFP–Atg13 in Atg13 knockout MEFs failed to rescue the formation of starvation-induced puncta by the endogenous early autophagy markers WIPI2 and Atg16, whereas expression of wt GFP–Atg13 did (Fig. 6D and supplementary material Fig. S5). We concluded that the positively charged residues at the N-terminus of Atg13 are important for the downstream formation of autophagic puncta.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Expression of mutant Atg13 inhibits formation of puncta by early autophagy markers. (A) HEK293 cells stably expressing the mutant Atg13 at different levels were starved, immunolabelled with antibodies against ULK1, Atg101, WIPI2 and Atg16, and their response was compared with the wt GFP–Atg13 clone. Representative images for Atg101 labelling are shown. (B) of The number of ULK1, Atg101, WIPI2 and Atg16 puncta normalized to the corresponding number of puncta of the wt clone 35, for five different fields containing 5–10 cells each (values are means ± s.d.). (C) Clonal cell lines stably expressing the wt and mutant GFP–Atg13 from Fig. 5D were starved and immunolabelled with antibodies against ULK1, Atg101 and WIPI2. Values are means ± s.d. of the number of ULK1, Atg101 and WIPI2 puncta normalized to the corresponding number of puncta of the wt clone 35, for five different fields containing 15–30 cells each. (D) Atg13 knockout (KO) MEFs stably expressing wt or mutant GFP–Atg13, were starved or treated with PP242 for 1 hour and immunolabelled for GFP and endogenous WIPI2. Scale bars: ∼10 µm.

Translocation of the ULK1 complex to early autophagic structures, including omegasomes, is stabilised by sustained PtdIns3P synthesis

We have found that when HEK293 cells are treated with wortmannin under autophagy-inducing conditions, the number of puncta formed by the ULK1 complex decreases. At the same time, live imaging experiments using GFP–Atg13 as a proxy for the ULK1 complex, show that wortmannin reduces the number of newly formed autophagic puncta, and those few that form have a much shorter lifespan. Thus, PtdIns3P synthesis is important for the recruitment of the ULK1 complex to autophagic puncta, and for its stabilization on these structures. This is very consistent with another of our findings, namely that the ULK1 complex becomes part of the omegasome structure, a signature membrane for PtdIns3P synthesis during autophagy.

Our findings that, in the absence of PtdIns3P synthesis, some ULK1 complex is recruited to puncta, are in broad agreement with a previous study of mouse embryonic fibroblasts (Itakura and Mizushima, 2010), and are consistent with the idea that translocation of ULK1 to a pre-autophagosomal membrane, most probably the ER, is a pre-requisite for the subsequent activation of the Vps34 complex (Di Bartolomeo et al., 2010; Matsunaga et al., 2010). It is possible that newly formed PtdIns3P may create a positive feedback loop, which reinforces the recruitment of the ULK1 complex, further stimulating PtdIns3P synthesis. This interplay between translocation of the ULK1 complex and PtdIns3P synthesis would be very rapid and may be the reason why we could not establish that during normal autophagy the formation of ULK1 puncta clearly precedes omegasome emergence. A recent study has shown that during autophagosome formation, Atg14, which is the autophagy-specific component of the Vps34 complex, translocates to the ER–mitochondria contact points (Hamasaki et al., 2013); the formation of these contact points is necessary for the correct localization of Atg14. An intriguing possibility is that PtdIns3P synthesis stabilizes the ER–mitochondria contact points, which feed back to the ULK1 complex and promote autophagosome formation.

ER is the most frequent proximal membrane to new autophagosomes

The very clear response to starvation of cells expressing GFP–Atg13, and the certainty by which newly formed puncta could be identified by spinning disk confocal microscopy, prompted us to systematically examine the spatial relationship of early autophagic structures to various cellular membranes. In all, we examined over 180 biogenesis events. We found that emerging autophagic puncta are distinct from the strands of the ER, but they remain constantly bound to the ER during their biogenesis. In contrast, the nature of the association between these particles and mitochondria is less tight and a similar result holds for lysosomes. These findings are true even when downstream expansion of the early GFP–Atg13 structures is blocked by wortmannin, and are consistent with previous work (Itakura and Mizushima, 2010). Since the ULK1 complex is the earliest functional complex of the core autophagy machinery that targets the precursor structures of autophagosomes, its localization provides insight into the origin of the autophagosomal membrane and the mechanisms by which it might be delivered. The stable association between the particles of the ULK1 complex and ER implies that the ULK1 complex targets a membrane compartment that is connected to ER. We can imagine two alternative types of connection: membrane continuity or peripheral binding. Membrane continuity suggests that membrane can be provided to the expanding isolation membrane (IM) directly. In contrast, peripheral binding suggests that membrane can be provided to the IM indirectly, possibly employing machinery similar to the mitochondrial association ER membranes (MAMs) (Kornmann et al., 2009). The nature of the association between the ULK1-positive IM and mitochondria, the second candidate source of the autophagosomal membrane (Hailey et al., 2010), is transient, precluding that membrane continuity between these organelles exists, at least as a general phenomenon. Therefore, membrane transport should be indirect, through vesicular carriers or membrane contact sites. Indeed, the MAMs could offer a plausible mechanism to transport membrane, either directly from mitochondria to autophagosomes or indirectly through ER, consistent with the fact that the ER is always in close proximity (Hamasaki et al., 2013). Finally, we can envision that membrane originating from the third candidate source, the plasma membrane, could be transported to autophagosomes through endocytic vesicles [such as those containing the ULK1 protein (Longatti and Tooze, 2009)] that fuse with the IM, in agreement with two recent studies (Moreau et al., 2011; Nair et al., 2011).

Relationship of the ULK1 complex with omegasomes and with the isolation membrane

Almost simultaneously to its translocation to ER, the ULK1 complex associates with one DFCP1-positive omegasome, within which one IM first appears. The nature of this association is not known. One plausible mode of interaction in trans is that the ULK1 punctum may comprise a cluster of vesicles instead of being a single structure [such as the ULK1-containing recycling endosomes shown by Longatti et al. (Longatti and Tooze, 2009)], but we cannot discern any such collection of vesicles. Our favoured interaction mode is that the particles of the ULK1 complex consist of the same membranes as the corresponding omegasomes and IMs. The ULK1 complex could target one or both sides of the flat IM, and the DFCP1 ring would mark the boundaries of the LC3-positive IM. This spatial arrangement would allow ULK1 to access and bind LC3, something that has been recently reported (Kraft et al., 2012). The driving force for the omegasome–ULK1 interaction may involve multiple proteins. It is clear from a number of studies that the ULK1 complex must have some affinity for membranes (Chan et al., 2009), especially those related to the ER (Chan et al., 2009; Itakura and Mizushima, 2010), since upon the induction of autophagy the early ULK1-positive puncta are found associated with these membranes. In yeast, where a partial crystal structure exists, it appears that a region of Atg1 (Atg1 is equivalent to the mammalian ULK proteins) contains a domain that responds to membrane curvature (Ragusa et al., 2012), and an equivalent region in ULK1 has also been reported (Chan et al., 2009). Based on our data, a second region that may allow interaction of the ULK1 complex with membranes is at the N-terminus of Atg13 and it depends on a cluster of positively charged residues. The dominant-negative effect of the mutations at the N-terminus of Atg13 on autophagy suggests that Atg13 may have an essential function in the translocation of the entire ULK1 complex to puncta, which would lead to autophagy induction. Indeed, we have shown that, in Atg13 knockout MEFs, the wt GFP–Atg13 could rescue the formation of WIPI2 and Atg16 puncta in contrast to the mutant. This finding suggests that the ULK1 complex requires the affinity of Atg13 for membranes in order to stimulate autophagosome formation.

We cannot at this point reach a conclusion as to the lipid preference of Atg13. For protein obtained from mammalian lysates, similar preference for PA, PtdIns3P and PtdIns4P was detected, whereas protein isolated from bacteria showed a stronger preference for PA, followed by PtdIns3P and lastly for PtdIns4P. In the absence of a crystal structure, the safest hypothesis is that Atg13 has affinity for negatively charged membranes (the three preferred lipids would all be the simplest representative of such a membrane), somewhat analogous to the situation with Atg14, another early autophagy protein driving the Vps34 complex to the ER (Fan et al., 2011; Matsunaga et al., 2010).

A model of autophagosome formation incorporating ULK1 and omegasomes

Aligning the findings of this study with the literature, we propose the following model for autophagosome formation through an omegasome intermediate (Fig. 7). Once the mTOR complex is inactivated, the ULK1 complex becomes activated and translocates to the precursors of omegasomes, which are associated with the ER. On the ER, the ULK1 complex contributes to the activation of the Vps34 complex synthesising PtdIns3P for the nucleation of omegasomes. PtdIns3P further stimulates the recruitment of the ULK1 complex, creating a positive feedback loop that sustains the activation of additional molecules of the Vps34 complex. The continuous synthesis of PtdIns3P drives the formation and expansion of the DFCP1-positive omegasomes and leads to the recruitment of downstream components. Such downstream components are the WIPI proteins and perhaps Atg9 vesicles, which are followed by the assembly of the ubiquitin-like conjugation systems and the lipidation of LC3. Having a system in place for the localised lipidation of LC3 will further induce the expansion of the IM, while the ULK1 complex remains attached to the IM. When this machinery has achieved expansion of the mixed membrane to its maximum size, the ULK1 complex dissociates from the IM, the omegasome starts collapsing, and the LC3-positive IM buds off from the omegasome ring. At the latest stage, the omegasome ring closes, while the mature autophagosome is released and eventually it fuses with the lysosomes. The mechanism by which the ULK1 complex dissociates from the early membranes before autophagosome completion is unknown. One possibility is that inhibition of PtdIns3P synthesis [or activation of its degradation by Jumpy (Vergne et al., 2009) or other PtdIns3P phosphatases that might localise on the IM (Taguchi-Atarashi et al., 2010)] might break the positive feedback loop that allows the complex to be stabilised on membranes. Alternatively, a phosphorylation–dephosphorylation on the ULK1 complex mechanism might be at play. This interesting question will be explored in future work.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

Model of autophagosome formation through an omegasome intermediate. The autophagosome formation cascade begins in the cytosol with a signal that inactivates mTOR (step 1). Then, the activated ULK1 complex translocates to ER, where it activates the Vps34 complex, inducing the synthesis of PtdIns3P (PI3P) and nucleating a new omegasome (step 2). Sustained synthesis of PtdIns3P creates of positive feedback loop that reinforces the recruitment of additional molecules of the ULK1 complex on the IM, and stimulates the expansion of the omegasome (step 3). At the same time, the ubiquitin-like conjugation systems assemble on the IM where they begin to lipidate LC3 molecules, further inducing the expansion of the IM (step 4). Next, the ULK1 complex dissociates from the IM, the omegasome ring begins to close, and LC3-positive membrane buds off of the omegasome ring (step 5). Finally, the omegasome collapses back to ER, the autophagosome is released (step 6) and eventually its lumen becomes acidic and fuses with the lysosomes for the degradation of its contents (step 7).