302 | Nature | Vol 578 | 13 February 2020
Article
(Fig. 1e) and coalesced to form a larger punctum (Extended Data Fig. 1g,
Supplementary Video 3). To observe the process of PAS formation in
detail, we optimized the expression level of GFP–Atg13 using the induc-
ible GAL1 promoter. When yeast cells accumulating GFP–Atg13 during
a 7-h induction were treated with rapamycin for 10 min, multiple small
GFP–Atg13 puncta appeared and coalesced to form a large punctum
(Fig. 1f, Supplementary Video 4). Upon coalescence of two puncta, the
aspect ratio changed from approximately 2.0 to 1.0 within a few seconds,
reflecting the liquid-like nature of puncta (Fig. 1g, Extended Data Fig. 1h,
Supplementary Video 5). We occasionally observed the enlargement
of one PAS punctum coinciding with the reduction of another (Fig. 1h,
i, Extended Data Fig. 1i, Supplementary Video 6). This phenomenon
is consistent with Ostwald ripening, although local phosphorylation
events might dissolve the shrinking PAS punctum. Collectively, these
data suggest that the starvation-induced PAS is a liquid-like biomo-
lecular condensate that is formed by liquid–liquid phase separation.
The Atg1 complex forms droplets in vitro
We previously reported that a higher-order assemblage of the Atg1
complex organizes the PAS and initiates autophagy^6. With the excep-
tion of Atg17, the components of the Atg1 complex contain many IDRs
(Fig. 2a, Extended Data Fig. 2a). We speculated that the higher-order
assemblage of the Atg1 complex induces liquid–liquid phase separa-
tion, providing a mechanism for formation of a liquid-like PAS. SNAP-
tagged Atg1, Atg13, and Atg17–Atg29–Atg31 were purified (Extended
Data Fig. 2b) and labelled with distinct fluorescent dyes. Mixing of
these proteins resulted in the immediate onset of phase separation,
indicated by the appearance of multiple spherical droplets, the number
of which reached a maximum at 15 min before a subsequent decline,
whereas the total area occupied by the droplets increased continu-
ously (Fig. 2b). Again, droplets coalesced to form a larger spherical
droplet, during which the aspect ratio changed from approximately 2.0
to 1.0, suggesting a liquid-like state (Fig. 2c, d, Extended Data Fig. 2c).
Droplets emitted three distinct fluorescence signals corresponding to
Atg1, Atg13 and Atg17–Atg29–Atg31, indicating that these components
colocalize within droplets (Fig. 2e). Components of biomolecular con-
densates can be divided into two qualitative classes: scaffolds, which
are essential for the formation of condensates, and clients, which are
dispensable^1. As Atg1 was dispensable for droplet formation (Extended
Data Fig. 2d, buffer), Atg1 can be considered as a client, whereas the
other components act as scaffolds; we therefore refer to the Atg13–
Atg17–Atg29–Atg31 droplets as scaffold droplets. These scaffold drop-
lets were promptly dispersed by 1,6-hexanediol treatment (Extended
Data Fig. 2d, e), similar to the PAS in vivo (Fig. 1e). Phase separation was010020001 .02.03.04.05.00 13
26301.01.52.0012340.2 00.40.60.81.00246810afbDIC0Before quenchAfter quench (s)
10GFP–Atg13Fluorescence recoveryTime (s)c00 .8 2.6Time (s)Aspect ratioDICGFP–Atg13 (s)DICGFP–Atg13 (min)
0132630Position (μm)FI (AU)Time (min)dghPAS Cytoplasmit1/2 = 1.3 ± 0.1 skoff = 0.53 ± 0.06 s–1FI (AU)
W (s) Position (μm)G(W) (×–3 10
)G(W) (×–3 10
)Beforequench After quench (s) 0 0.71,6-hexanediol
3 min Removal1 mineDICGFPRapa (3 h)Addition^046171183479DICGFP–Atg13 (s)
1370100200
0120 12012PAS CytoplasmD (μm2 s–1)02468100
10 –4 10 –2 100 102 10 –4 10 –2W (s) 100 1020.250.500.751.001.2500.250.500.75Fig. 1 | The PAS behaves as a liquid droplet in vivo. a, Left, rapid recovery of
fluorescence of GFP–Atg13 puncta after photobleaching. Right, fluorescence
recovery fitted to a curve; dissociation rate constant (koff) and recovery
half-time (t1/2) values (mean ± s.d.) were calculated from n = 3 independent
experiments. DIC, differential interference contrast microscopy. b, FCS
measurements of GFP-Atg13 in the PAS and in the cytoplasm. The
autocorrelation function of the f luorescence signal is calculated for a single
pixel near the centre of the PAS, or for the five pixels in the cytoplasm (insets).
The autocorrelogram was fitted to a three-dimensional one-component
diffusion model to estimate diffusion coefficients. c, Tukey-style box-and-
whisker plot^28 of the diffusion constants of GFP–Atg13 in the PAS (n = 35 cells)
and in the cytoplasm (n = 32 cells) measured by scanning FCS (see Methods for
details). The 95% confidence interval of the medians are shown with notches,
and are estimated to be 0.79 ± 0.23 μm^2 s−1 (PAS) and 0.93 ± 0.27 μm^2 s−1
(cy toplasm). *P = 0.031, two-sided Wilcoxon Mann–Whitney U-test. d, Partial
FR AP experiment with a giant PAS. Graphs indicate line profiles of f luorescence
intensity (FI) in the above images. e, Reversible effect of 1,6-hexanediol on the
formation of Atg13–GFP puncta. Rapa, rapamycin treatment. f, Appearance of
multiple Atg13–GFP puncta and their coalescence to form one large punctum.
g, Left, coalescence of two PAS precursors. Right, graph shows the change in
aspect ratio during coalescence. h, Ostwald ripening of PAS precursors
observed in vivo. i, Line profile of f luorescence intensity in h. All scale bars are
2 μm. Experiments were repeated independently twice (e, f) or three times
(a, d, g, h) with similar results.
0246810120204060801000510 15 20Number Average area010203040506070Atg1Atg13Atg17Atg29Atg31
ca bAtg17DICMerge0 8 Time (s) 83 145Time (min)Number of dropletsAtg1 Atg13 Atg17 DIC Merge01020Time (min)d efTime (s)Aspect ratioAtg13
Atg17WT
WTF375A F430A
D247AP393AWT WT
WT WTAverage area (AU)Atg1317BRAtg17Atg17F430 D247D247Atg1317LRAtg17Atg17P393P393
F375g1.01.21.41.61.82.0050 100 150Predicted disorderedresidues (%)Fig. 2 | The Atg1 complex undergoes phase separation in vitro. a, Percentage
of residues predicted to be disordered by DISOPRED^29. b, Formation of liquid
droplets of the Atg1 complex on mixing. The bottom graph shows the
mean ± s.d. of number and area of the droplets (n = 3 independent
experiments). Scale bar, 50 μm. c, Coalescence observed between the droplets
of the Atg1 complex. Scale bar, 5 μm. d, Time course of the changes in aspect
ratio during coalescence in c. e, Co-occurrence of Atg1, Atg13 and Atg17 in the
Atg1-complex droplets. Scale bar, 30 μm. f, Interactions of Atg13 17BR (left)
and 17LR (right) with Atg17 (PDB 5JHF). Residue numbers refer to the
Saccharomyces cerevisiae Atg17 sequence. g, Formation of scaffold-complex
liquid droplets and their inhibition by mutations. Scale bar, 50 μm.
Experiments were repeated independently three times with similar
results (c, e, g).