Nature | Vol 578 | 13 February 2020 | 301Article
Phase separation organizes the site of
autophagosome formation
Yuko Fujioka^1 , Jahangir Md. Alam^1 , Daisuke Noshiro1,2, Kazunari Mouri^3 , Toshio Ando^2 ,
Yasushi Okada3,4, Alexander I. May5,6, Roland L. Knorr7,8,9, Kuninori Suzuki1 0,1 1,
Yoshinori Ohsumi^5 & Nobuo N. Noda^1 *Many biomolecules undergo liquid–liquid phase separation to form liquid-like
condensates that mediate diverse cellular functions^1 ,^2. Autophagy is able to degrade
such condensates using autophagosomes—double-membrane structures that are
synthesized de novo at the pre-autophagosomal structure (PAS) in yeast^3 –^5. Whereas
Atg proteins that associate with the PAS have been characterized, the
physicochemical and functional properties of the PAS remain unclear owing to its
small size and fragility. Here we show that the PAS is in fact a liquid-like condensate of
Atg proteins. The autophagy-initiating Atg1 complex undergoes phase separation to
form liquid droplets in vitro, and point mutations or phosphorylation that
inhibit phase separation impair PAS formation in vivo. In vitro experiments show that
Atg1-complex droplets can be tethered to membranes via specific protein–protein
interactions, explaining the vacuolar membrane localization of the PAS in vivo. We
propose that phase separation has a critical, active role in autophagy, whereby it
organizes the autophagy machinery at the PAS.The PAS is a transient structure that is regulated by nutrient conditions
and invariably forms on the vacuole in yeast on starvation^3. The PAS ini-
tially comprises Atg1 complexes consisting of Atg1, Atg13, Atg17, Atg29
and Atg31, which are abundant with intrinsically disordered regions
(IDRs)^6. This ‘early PAS’ then matures by recruiting downstream Atg pro-
teins and vesicles, subsequently serving as the site of autophagosome
formation^7 ,^8. These features are consistent with biomolecular conden-
sates (also known as membraneless organelles) that are formed through
liquid–liquid phase separation^1 ,^2. We therefore set out to determine
whether the PAS is in fact a biomolecular condensate, and examined
whether condensate formation is able to explain its spatiotemporal
behaviour in the cell.
The PAS is a liquid-like condensate
First, we studied the dynamics of the PAS using fluorescence micros-
copy. We used yeast cells overexpressing GFP–Atg13 in an atg11Δ
background to ensure that fluorescence intensity is sufficient for
quantitative analysis and that the PAS is formed in response to starva-
tion rather than the Atg11-dependent pathway, which is constitutive
and responsible for cytoplasm-to-vacuole targeting^9 ,^10. We confirmed
that overexpression of GFP–Atg13 did not impair autophagy activity
(Extended Data Fig. 1a). Upon nitrogen starvation, GFP–Atg13 formed
puncta that dissolved within 8 min of addition of a nitrogen source
(Extended Data Fig. 1b), consistent with a previous study^9 , suggesting
that the PAS is a dynamic and transient entity. Fluorescence recovery
after photobleaching (FRAP) experiments showed a quick recovery of
fluorescence in puncta (corresponding to the PAS) containing GFP–
Atg13, with a recovery half-time of 1.3 s after photobleaching (Fig. 1a,
Extended Data Fig. 1c, Supplementary Video 1). This exchange rate
is comparable with or even faster than that of molecules in nuclear
biomolecular condensates^11. We also observed rapid fluorescence
recovery of Atg1–GFP, Atg13–GFP and Atg17–GFP proteins expressed
at endogenous levels (Extended Data Fig. 1d). We next performed flu-
orescence-correlation microscopy (FCS) experiments on GFP–Atg13.
The diffusion coefficient of GFP–Atg13 in the PAS was about half that
in the cytosol (Fig. 1b, c, Extended Data Fig. 1e). These results provide
evidence supporting a dynamic liquid-like structure of the PAS in which
GFP–Atg13 can move diffusively, although the crowded environment of
the PAS would have slowed down the diffusion, as reported with other
liquid-droplet structures^12 ,^13. Owing to the small size of the PAS, the rate
determined by FCS could potentially include both entry and exit of GFP–
Atg13 molecules to the PAS. We next generated and performed FRAP
analyses on a giant PAS (diameter >1 μm) by overexpressing GFP–Atg13
from a multicopy plasmid; Atg13–GFP fluorescence rapidly recovered
(less than 1 s) in a partially quenched region (Fig. 1d, Extended Data
Fig. 1f, Supplementary Video 2). These data show that Atg13 is not only
able to enter and exit the PAS, but also moves freely within the PAS.
Previous studies have established that 1,6-hexanediol is able to inhibit
liquid–liquid phase separation of biomolecules^14. When yeast cells were
treated with 1,6-hexanediol, Atg13 puncta rapidly dissolved; they reap-
peared on the vacuolar membrane after removal of 1,6-hexanediolhttps://doi.org/10.1038/s41586-020-1977-6
Received: 17 December 2018
Accepted: 2 December 2019
Published online: 5 February 2020
(^1) Institute of Microbial Chemistry (BIKAKEN), Tokyo, Japan. (^2) Nano Life Science Institute (WPI-NanoLSI), Kanazawa University, Kanazawa, Japan. (^3) Center for Biosystems Dynamics Research (BDR),
RIKEN, Osaka, Japan.^4 Department of Physics, Universal Biology Institute (UBI) and International Research Center for Neurointelligence (WPI-IRCN), The University of Tokyo, Tokyo, Japan.^5 Cell
Biology Center, Institute of Innovative Research, Tokyo Institute of Technology, Yokohama, Japan.^6 Tokyo Tech World Research Hub Initiative (WRHI), Institute of Innovative Research, Tokyo
Institute of Technology, Yokohama, Japan.^7 Department of Theory and Bio-Systems, Max Planck Institute of Colloids and Interfaces, Potsdam, Germany.^8 Graduate School and Faculty of
Medicine, The University of Tokyo, Tokyo, Japan.^9 Max Planck Institute of Molecular Plant Physiology, Potsdam, Germany.^10 Life Science Data Research Center, Graduate School of Frontier
Sciences, The University of Tokyo, Kashiwa, Japan.^11 Collaborative Research Institute for Innovative Microbiology, The University of Tokyo, Tokyo, Japan. *e-mail: [email protected]