304 | Nature | Vol 578 | 13 February 2020
Article
throughout these droplets, confirming that droplets are not regularly
structured (Fig. 4a). S-shaped Atg17 molecules exhibited dynamic
behaviour in a restricted area within the droplets (Supplementary
Video 7), providing further evidence that the droplets are in a liquid-
like state. By contrast, Atg17 was arranged in a regular pattern and
showed little movement in droplets that were loaded onto a positively
charged coverslip (Fig. 4b, Supplementary Video 8), suggesting that
the droplets can mature to a static, solid-like structure, depending on
the environment. When FRAP experiments were performed on scaffold
droplets, we observed that photobleaching impaired coalescence of
droplets and that Atg17 fluorescence rarely recovered (Extended Data
Fig. 4). Collectively, these observations indicate that the droplets are
liquid-like and randomly structured, and that the droplets eventually
mature into a static, ordered state in vitro, as has been observed in
other biomolecular condensates that transition from liquid droplets
to solid-like states such as gels, glasses and amyloids^24.
In vitro reconstitution of the early PAS
PAS formation occurs on the vacuolar membrane. Previous studies
indicate that the autophagy-related vacuolar membrane protein Vac8
interacts directly with Atg13^25. Monitoring of GFP–Atg1, Atg5–GFP and
GFP–Atg8 during starvation showed that puncta containing these
proteins were mostly attached to the vacuolar membrane in wild-type
cells, whereas about half were detached from the vacuolar membrane
in vac8Δ cells (Fig. 5a, b, Extended Data Fig. 5a, b). This suggests that
Vac8 is at least partly responsible for tethering the PAS to the vacuolar
membrane. We next investigated whether Atg1-complex droplets could
also be tethered to membranes via Vac8 using giant unilamellar vesicles
(GUVs). Atg1-complex droplets were tethered to Vac8-anchored GUVs
but not to GUVs lacking Vac8 (Fig. 5c, d, Extended Data Fig. 5c); Atg1
was dispensable for droplet binding to Vac8 GUVs (Fig. 5e). The scaffold
droplets were only rarely tethered to GUVs in the presence of the Vac8
mutant that lost the affinity with Atg13^26 , further confirming that tether-
ing occurs through a specific Atg13–Vac8 interaction (Fig. 5f, Extended
Data Fig. 5d, e). The number of droplets tethered to wild-type Vac8 GUVs
was rapidly reduced, while their size increased through coalescence
events (Fig. 5g, h, Extended Data Fig. 5f, Supplementary Video 9). This
result is consistent with our in vivo and in vitro findings (Fig. 1 f, g, 2c),
revealing that the droplets remain in a liquid-like state, even in GUVs.
FRAP experiments revealed that Atg1-complex droplets tethered to the
membrane exchanged 60–100% of their constituent Atg1 molecules
within 3 min (Extended Data Fig. 5g), providing further evidence for
the liquid-like nature of droplets, even when on membranes. On the
basis of these observations, we conclude that a structure similar to
the early PAS was reconstituted in vitro using purified proteins and
synthetic liposomes, demonstrating that the early PAS is a liquid-like
condensate that is tethered to the vacuolar membrane through a spe-
cific protein–protein interaction.Discussion
The relationship between biomolecular condensates and autophagy
is generally thought of in passive terms: condensates are targeted
for degradation by autophagy^5. Our results challenge this notion,
instead suggesting that the PAS—the central driver of the autophagy
mechanism—is a liquid-like biomolecular condensate (summarized in
Extended Data Fig. 6). Phase separation is implicated at a fundamental
level in PAS formation, with the interactions between IDR-containing
Atg1-complex components critical for the early PAS. In previous work,
extensive structural analyses have been performed on the Atg1 complex
that have established the stoichiometry of Atg1–Atg13, Atg13–Atg17
and Atg17–Atg29–Atg31 interactions^16 ,^23. However, in these studies, the
majority of IDRs were removed for technical reasons, preventing the
observation of Atg1-complex phase-separation events. Here we reveal
that cross-linking of Atg17 dimers by the long IDR of Atg13 is the main
mechanism of phase separation, a finding supported by a previous
structural study describing specific and multivalent Atg13–Atg17 inter-
actions^6. The resulting liquidity of the PAS is critical for its function in
dynamic recruitment of Atg proteins throughout autophagosome for-
mation: for example, liquidity probably results in the concentration and
activation of Atg1 kinase for autophagy initiation, and would facilitate
the incorporation of Atg9 vesicles, the initial source of autophagosomal
membranes^8 , in a manner reminiscent of liquid-phase synapsin cluster-
ing of vesicles at synapses^27. For these functions, the liquidity of the
PAS—which is easily lost by maturation, as observed in vitro (Fig. 4 )—is
maintained in cells through formation and dissolution events that are
mediated by a combination of kinases and phosphatases.abFluorescence HS-AFM imageAtg17 Atg17Atg13 Atg130 nm90 nm 300 nm0 nmFluorescence HS-AFM imageFig. 4 | The surface structure of scaffold droplets is irregular and dynamic.
a, b, Visualization of S-shaped Atg17 molecules in scaffold droplets on non-
coated (a) and 3-aminopropyltriethoxysilane-coated (b) coverslips using HS-
AFM in combination with f luorescence microscopy. Broken lines indicate the
position of the cantilever used for HS-AFM observation. Inset shows Atg17
dimer structure (PDB 5JHF) in the same scale. Bottom HS-AFM images are after
fast Fourier transform-bandpass filter. Experiments were repeated
independently three times with similar results (a, b). a, b, Scale bars, 5 μm
(left), 100 nm (a, right) and 200 nm (b, right).