Pexophagy Protects Plants from Reactive Oxygen Species-induced Damage under High-intensity Light


 Light is essential for photosynthesis, but it has the potential to elevate intracellular levels of reactive oxygen species (ROS) during photosynthesis. Photorespiration is a metabolic pathway in photosynthesis to metabolise oxidised by products from chloroplasts, and it generates a high level of ROS in peroxisomes. Since high levels of ROS are toxic, plants must manage damage from ROS. However, the cellular mechanism to elude leaf damage from ROS in peroxisomes is not fully explored. Here we show that autophagy plays a pivotal role in the selective removal of ROS-generating peroxisomes, which protects plants from oxidative damage during photosynthesis. We found that a series of peup mutants, which is a defect in autophagy degradation of peroxisomes, showed light-intensity-dependent leaf damage and excess aggregation of ROS-accumulating peroxisomes. The peroxisome aggregates were specifically engulfed by pre-autophagosomal structures and vacuolar membranes, but they were not degraded in the mutants. ATG18a-GFP and GFP-2 × FYVE, which both bind to phosphatidylinositol 3-phosphate, preferentially targeted the peroxisomal membranes and pre-autophagosomal structures near peroxisomes in ROS-accumulated cells under high-intensity light conditions. Our findings provide new information to better understand the plant stress response caused by light irradiation.


Introduction
Photosynthesis in plants converts light energy to chemical energy and is accompanied by photorespiration, which involves peroxisomes, mitochondria, and chloroplasts 1 . Photorespiration is essential for plant survival under high-intensity light to prevent photoinhibition, which damages photosynthetic machinery owing to excess reactive oxygen species (ROS) accumulation 2, 3 . Thus, understanding how excess ROS is degenerated to protect plants from oxidative damage during photosynthesis under excess light is essential.
A set of autophagy (ATG) genes are included in the speci c degradation of peroxisomes, namely pexophagy in yeasts and animals [15][16][17][18][19] . ATG proteins for the initial of autophagy form preautophagosomal structures (PAS) on vacuolar membranes containing phosphatidylinositol 3-phosphate (PtdIns3P) adjacent to degraded peroxisomes in yeast [20][21][22] . Subsequently, a membrane structure called the phagophore extends from the PAS to cover peroxisomes by incorporating phosphatidylethanolamine (PE) conjugated ATG8 (ATG8-PE), and then fuses to vacuoles for degradation 18,19,22 . Most ATG proteins are highly conserved in yeasts, animals, and plants [15][16][17][18][19] . However, it is unclear whether pexophagy in plants is the same as in yeasts and animals, because homologues of key factors for pexophagy in yeasts, namely PpAtg30 and ScAtg36, are absent in plants. Moreover, no direct evidence exists for selective degradation of peroxisomes by pexophagosomes in plant cells 23 .
Here, we investigate the cell-structural mechanism for autophagy-dependent degradation of ROSaccumulated peroxisomes to capture the pexophagosome formation in Arabidopsis leaves. Furthermore, we examined the impact of the de ciency of pexophagy on the leaf damage caused by ROS accumulation, which is enhanced by the accumulation of inactive catalases in peroxisomes. Our nding indicates a massive contribution of pexophagy on protecting plants from excess light-induced oxidative damage during photosynthesis.
ATG18a preferentially targets leaf peroxisomes in light-adapted cells ATG18a plays a role in autophagosome formation to degrade oxidative proteins in Arabidopsis 27 , implying that ATG18a targets damaged organelles. To examine whether ATG18a is required for selective pexophagy in plants, we assessed the intracellular distribution of ATG18a-GFP in wild-type plants and peup1/atg2 and peup4/atg7 mutants bearing red uorescence protein-fused peroxisomal targeting signal 1 (RFP-PTS1) to visualise peroxisomes in these plants (Fig. 2a). The ATG18a-GFP-containing structures localised to peroxisomes, although they are rarely observed in the wild-type plants ( Fig. 2a-d). We found that many cells accumulated ATG18a-GFP structures on peroxisomes in the peup1/atg2 and peup4/atg7 mutants (Fig. 2b) and found that 30-40 % of the total peroxisomes, especially aggregated peroxisomes, had the ATG18a-GFP structures (Fig. 2c). Furthermore, about 80 % of the ATG18a-GFP structures localised to peroxisomes in wild-type and mutant cells (Fig. 2d), suggesting that ATG18a preferentially targets the peroxisome. Immunoblot showed ATG18a-GFP and catalase in the insoluble fraction of peup1/atg2 ( Supplementary Fig. 5). Most ATG18a-GFP appeared as dot structures, but a few were cup or ring structures in peup1/atg2 and peup4/atg7 mutants at 100 µmol m -2 s -1 light intensity (Fig. 2a, e-g and Supplementary Fig. 6a, c and Supplementary Table 1). We tracked single peroxisomes by time-lapse imaging, and then the average image of RFP-PTS1 and ATG18a-GFP ( Fig. 2h) was generated using a morphological image processing tool 28 . The image revealed a ring structure of ATG18a-GFP surrounding the peroxisome in peup4. Time-lapse imaging also showed that ATG18a-GFP gradually surrounded peroxisomes in wild type and peup4/atg7, but not in peup1/atg2 ( Supplementary Fig. 7a-c and Supplementary Videos 5-7). Indeed, peup1/atg2 had fewer ring structures compared with peup4/atg7 (Fig. 2f). In the peup1/atg2 mutant, 60 % of uorescence from ATG18a-GFP was recovered within 60 s after photobleaching, indicating that ATG18a-GFP rapidly accumulates at the peroxisome aggregates (Supplementary Fig. 8 and Supplementary Video 8).
To examine whether ATG18a interacts with other proteins, we conducted an immunoprecipitation of ATG18a-GFP followed by protein mass spectrometry. The result shows that various proteins of chloroplasts, peroxisome, and mitochondria were co-immunoprecipitated with ATG18a-GFP in peup1/atg2 (Supplementary Fig. 9 and Supplementary Table 2). Peroxisome proteins such as catalases (CAT1, CAT2 and CAT3), heat shock protein 70s (HSP70s), and RuBisCO-related proteins were abundantly present. We obtained the number of proteins localised to each organelle from two databases (PPDB, http://ppdb.tc.cornell.edu/dbsearch/subproteome.aspx; and SUBA4, http://suba.live) and calculated the recovery rate. The peroxisome proteins were more e ciently recovered compared with the mitochondria and chloroplast proteins ( Supplementary Fig. 9b), suggesting that many ATG18a proteins directly or indirectly bind to peroxisomes or peroxisomal proteins.
In detailed analysis with an electron microscope, we observed the ER and autophagosome-like structures adjacent to the high-density area in peroxisomes of peup1/atg2 mutants ( Supplementary Fig. 11b, c). We further examined the localisation of GFP-2×FYVE and ATG18a-GFP in detail with immunoelectron microscopic analysis using anti-GFP antibodies ( Supplementary Fig. 11b). The results show that GFP-2×FYVE and ATG18a-GFP were localised on both peroxisomes and phagophores adjacent to the peroxisomes, suggesting that ATG18a-GFP recognises PtdIns3P on the membrane of damaged peroxisomes or PAS to initiate pexophagy in plants.
Remarkably, the large aggregates of peroxisomes were induced in leaf mesophyll cells of atg2, atg5, and atg7, mostly at the cell bottom ( Fig. 4b and Supplementary Fig. 13b). the frequency and size of peroxisome aggregation in atg2, atg5, and atg7 under high-intensity light conditions was two to three times greater than under normal light (100 µmol m -2 s -1 ) (Fig. 4c, d and Supplementary Figs. 13c, 14a, b). Peroxisome aggregation in atg2 and peup4/atg7 was evident ( Fig. 4b-d and Supplementary Fig. 13b, c). The accumulation of catalase was higher in the insoluble fractions of atg2, atg5, and peup4/atg7 (Fig.  4e, f and Supplementary Figs. 13d, 14c). We found an increase in the number of mitochondria, which gathered in peroxisome aggregation in peup4/atg7 under high-intensity light, suggesting that mitophagy was also suppressed (Supplementary Fig. 15a-c). Mitochondrial proteins serine hydroxymethyltransferase (SHMT) and cytochrome c oxidase 2 (COXII) were increased in peup4/atg7 under high-intensity light (Supplementary Fig. 15d-g). These results suggest that ATG7 plays multiple roles in the degradation of damaged mitochondria as well as peroxisomes in leaves undergoing photosynthesis.
High-intensity light-induced peroxisome aggregates are surrounded by vacuolar membranes together with ATG18a and PtdIns3P To further investigate whether high-intensity light-induced large aggregates of peroxisomes are degraded by autophagy, we focused on the subcellular localisation of ATG18a-GFP and GFP-2×FYVE in 1000 µmol m -2 s -1 light-adapted leaf cells in peup1/atg2 and peup4/atg7 mutants ( Fig. 5a-d). The results show that ATG18a-GFP and GFP-2×FYVE preferentially targeted the large aggregates of peroxisomes in peup1/atg2 and peup4/atg7 mutants. Especially, peroxisome aggregations in peup4/atg7 mutants were mostly enveloped by ATG18a-GFP (Fig. 5a, b and Supplementary Table 2); the frequency of these peroxisome aggregations was about 43 % in peup4/atg7, whereas it was about 11 % in peup1/atg2 ( Fig. 5c and Supplementary Table 4). The size of peroxisome aggregates enveloped by ATG18a-GFP in peup4/atg7 was about 34 µm 2 , which was eight times greater than that in wild type and six times than that in peup1/atg2 (Fig. 5d). In contrast, both the frequency and size of peroxisome aggregates enveloped by GFP-2×FYVE were smaller than by ATG18a-GFP in all tested lines (Fig. 5c, d). The analysis of uorescent intensities in the aggregates con rmed that ATG18a-GFP co-localised with the large aggregates of peroxisomes (Fig. 5e).
We further investigated the relationship between vacuolar membranes and peroxisomes in wild type and peup4/atg7 using Venus-VAM3 under high-intensity light. We found that vacuolar membrane structures surrounded peroxisome aggregates in peup4/atg7 (Fig. 5f, g). The frequency of these structures was similar to that of aggregate formations with ATG18a-GFP (Fig. 5c, h and Supplementary Tables 4, 6), suggesting a similar mechanism in these two structures. The cells with these vacuolar membrane structures were three times more abundant in peup4/atg7 than in wild type (Fig. 5h). The size of the peroxisome aggregation with the vacuolar membrane was also bigger in peup4/atg7 than in wild type and peup1/atg2 (Fig. 5i). Although the number of vacuolar bulbs was more abundant in peup4/atg7 than in wild type, the ratio of bulbs surrounding peroxisomes was similar between peup4/agt7 and wild type (Supplementary Fig. 16 and Supplementary Videos 11, 12).
High-intensity light-induced peroxisome aggregates accumulate ROS We investigated the accumulation of ROS in leaves under high-intensity light (1000 µmol m -2 s -1 ). NBT staining showed that the leaves of both peup1/agt2 and peup4/atg7 mutants accumulated more ROS compared with wild type (Supplementary Fig. 17a, b). Next, we examined the accumulation of ROS in peroxisomes using 2 7 -dichlorodihydro uorescein diacetate (H 2 -DCF-DA) 33,34 (Fig. 6a, b and Supplementary Fig. 17c-e). We found that some peroxisomes in wild type and about 60 % of peroxisome aggregation in the mutants were speci cally stained with H 2 -DCF ( Supplementary Fig. 17c, d). H 2 -DCF uorescence was detected inside peroxisomes in the mutants with approximately two-fold higher intensity than in wild type (Fig. 6b). The uorescence intensity in peroxisomes was two-to three-fold higher compared with chloroplasts; this was especially prominent in the mutants (Supplementary Fig.  17e). We concluded that ROS accumulates at high levels in peroxisome aggregates in peup1/atg2 and peup4/atg7 mutants grown under high-intensity light.
These results suggest that high-intensity light induces accumulation of high levels of ROS in peroxisomes, resulting in aggregation and subsequent degradation of peroxisomes in the vacuole by pexophagy.

Discussion
Autophagy preferentially degrades ROS-accumulated peroxisomes in light We found leaf damage in peup1/atg2 and peup4/atg7 mutants in light ( Supplementary Fig. 1). Increases in light intensity increased leaf damage, suggesting the involvement of photosynthesis. Because highintensity light induces ROS accumulation from photosynthesis, we speculated that light-induced ROS accumulation caused leaf damage in the mutants. Indeed, we found light-dependent ROS accumulation in the leaves of peup1/atg2 and peup4/atg7 mutants, indicating that these mutants generate higher levels of ROS compared with wild type under high-intensity light. These results suggest that high levels of ROS in peup4/atg7 induce the formation of peroxules and stromules ( Supplementary Fig. 4) 25,26 .
Autophagy is required for the degradation of damaged and toxic materials generated by ROS accumulation during oxidative stress 13 . However, the primary origin of ROS in leaf mesophyll cells of the autophagy-de cient mutants is still unclear. We hypothesised that the undegraded peroxisomes would primarily produce ROS in the mutants during metabolism in photorespiration. Indeed, hydrogen peroxide accumulation is higher in peroxisomes than in chloroplasts and mitochondria during photorespiration 35 . The H2-DCF-stained aggregates of peroxisomes in the mutants con rmed the accumulation of ROS in degrading peroxisomes (Fig. 6a, b and Supplementary Fig. 17c, d). Hydrogen peroxide in peroxisomes is immediately degraded by catalase in wild-type plants. However, catalase is gradually inactivated by increasing levels of ROS in photosynthetic tissues under high-intensity light conditions. The inactivation of catalase causes over-accumulation of ROS in peroxisomes and then induces the imbalance of ROS homeostasis in cells, leading to damage and defective plant growth in the mutants [4][5][6] . Peroxisome participates in photorespiration through physical interaction with chloroplast and mitochondrion 36 .
Therefore, damaged peroxisome with high ROS levels should be immediately removed by pexophagy to maintain e cient metabolite ow among these organelles during photorespiration under high-intensity light conditions.
We focused on ATG18a, which is involved in the degradation of oxidative proteins 27 , to assess how autophagy degrades peroxisomes. Because ATG18a has well conserved-PtdIns3P-binding domain in yeast, plant, and animals 21,22,29,37,38 , we used GFP-2×FYVE to monitor PtdIns3P in the cell. Both GFP-2×FYVE and ATG18a-GFP preferred to target peroxisomal aggregation in peup1/atg2 and peup4/atg7 under normal light (100 µmol m -2 s -1 ) (Figs. 2 and 3). Furthermore, we showed that high-intensity light (1000 µmol m -2 s -1 ) increased the frequency and the size of peroxisome aggregates in peup1/atg2, atg5, and peup4/atg7 mutants (Fig. 4b-d and Supplementary Figs. 13b, c, and 14a, b), with an increase in GFP-2×FYVE and ATG18a-GFP targeting ( Fig. 5c and Supplementary Table 5). These proteins form the autophagosome-like cup and ring structures that surround peroxisomes. These ndings indicate that the light-induced peroxisome aggregates are speci cally degraded via pexophagy. The peroxisomal aggregation in peup1/atg2 consists of oxidative peroxisomes with inactive catalase 11 . Therefore, ATG18a recognises the oxidative peroxisomes through binding activity with PtdIns3P to degrade them.
ATG18a-GFP was occasionally localised to places other than peroxisomes (Fig. 2d), such as chloroplasts ( Supplementary Fig. 18a Fig. 18a-e). This is consistent with previous reports showing that high-intensity light induces ROS accumulation in chloroplasts 26 , and subsequent degradation of damaged chloroplasts by autophagy (chlorophagy) 39 . Meanwhile, the relative intensity of H2-DCF from peroxisomes in peup1/atg2 and peup4/atg7 was about three times stronger than that from chloroplasts ( Supplementary Fig. 17c, e). We also found that autophagy contributed slightly to the degradation of mitochondria (mitophagy), but to a lesser degree than pexophagy under light ( Supplementary Fig. 15). We noticed that HSP70s were recovered in the pull-down assay of ATG18a-GFP (Supplementary Fig. 9 and Supplementary Table 2), implying the involvement of chaperone-mediated autophagy or microautophagy 40 . Collectively, these ndings suggest that various types of cellular components, mostly damaged peroxisomes, are degraded by autophagy under light.
Plants have a unique mechanism for pexophagy Selective autophagy has been well studied in yeast 15,18,22 and mammals 23,41,42 , but less so in plants 13,17 . The subcellular location of PtdIns3P synthesis during autophagy differs depending on the organisms and organelles to be degraded (e.g., vacuoles in yeast and omegasomes in mammals) [42][43][44][45][46] . In plants, the location of PtdIns3P synthesis, the origin of isolation membranes, and how ATGs participate in pexophagosome formation are unknown 13,17,47 .
We showed that many dot structures of ATG18-GFP (Fig. 2a, e, f) and GFP-2×FYVE (Fig. 3a, e, f) localise to peroxisomes in peup1/atg2 and peup4/atg7, suggesting that PtdIns3P is formed adjacent to the peroxisomes to attract ATG18a before the action of ATG2 and ATG7. The detailed analysis by electron microscope revealed that PtdIns3P and ATG18a were localised on both peroxisomes and phagophores adjacent to peroxisomes (Supplementary Fig. 11b). In wild type and peup4/atg7, we observed that a dot structure of ATG18-GFP or GFP-2×FYVE gradually change to a ring structure via a cup structure to engulf peroxisomes, but this change was not observed in peup1/atg2 (Supplementary Figs. 7, 10 and Supplementary Videos 5,6,7,9,10). The aggregated peroxisomes were captured in invagination into vacuoles in peup4/atg7 (Fig 5f, g, and Supplementary Fig. 16). Based on these results, we propose the following model for pexophagy (Fig. 6c): 1) peroxisomes with damaged catalase accumulate high levels of ROS, and PtdIns3P is generated on the peroxisome membrane or phagophores formed adjacent to peroxisome and ER, 2) ATG18a targets the PtdIns3P on the damaged peroxisomes, 3) pexophagosome is formed based on ATG18a and PtdIns3P with other autophagy factors, 4) pexophagosomes completely sequester damaged peroxisomes, and 5) pexophagosomes are incorporated into the vacuole.
We speculate that ROS generation is responsible for the induction of pexophagy, but it is still unclear how ROS generated in the peroxisome matrix are recognised for pexophagy. In human pexophagy, ataxiatelangiectasia mutated protein on the peroxisomal membrane senses ROS inside the peroxisome to induce pexophagy by mediating mTORC1 suppression and peroxin 5 (PEX5) phosphorylation 12,48 . Plant pexophagy might also involve sensor protein(s) along with plant PEX proteins on the peroxisome membrane to induce pexophagy 12,22,48 . In yeasts, receptors such as PpAtg30 and ScAtg36 interact with PEX3 and PEX14 to recognise peroxisomes to be degraded in pexophagy, but orthologues of these receptors are not found in plants 12,22,23,45,[47][48][49] . Alternatively, oxidised lipids on the peroxisome membrane may be the signal to induce pexophagosome formation, because they are the hallmark of oxidised peroxisomes. The accumulation mechanisms of PtdIns3P exist on both peroxisomes and phagophores. This is supported by the fact that multiple pathways for the accumulation of PtdIns3P are activated in autophagy 42,44,50 . In mitophagy in mammalian cells, activation of phosphoinositide 3kinase and inactivation of PTEN, a PtdIns3P phosphatase, occur on the membrane of initial phagophores, namely omegasomes 37,38,41,43 , which are derived from the ER as platforms executing mitophagy 51,52 . Recent studies have shown that phagophores in mammalian cells are generated from the contact site between the ER and mitochondria 37,38,41,43,53 and in plant cells from the ER in which ATG5, ATG9, and ATG18 are localised 54,55 . In yeast Saccharomyces cerevisiae, ATG2-ATG18 complex tethers PAS to ER for extending isolation membrane 56 . We showed that the ER and phagophores were located adjacent to the high-density area in peroxisomes of peup1/atg2 ( Supplementary Fig. 11b, c) and atg5 mutants 14 . ATG18a and PI(3)P were localized to the area (Figs. 2, 3). These ndings suggest that the initial phagophore generates at the site where the ER overlaps with a speci c receptor and the PtdIns3P on peroxisomes in plant pexophagy, acting as a platform of PAS (Fig. 6c). ATG18 gathers on the PtdIns3P for extension of pexophagosomes with a lateral supply of isolation membrane from ER.
Leaf damage and peroxisome aggregation in atg9 are reduced compared with those in atg2, atg5, and atg7 under high-intensity ( Supplementary Fig. 13) and normal light conditions 11,14 , suggesting that the contribution of ATG9 in plant pexophagy is small, unlike in yeast and mammal pexophagy 18,19,45,46 . ATG9 might not have a speci c role in pexophagy, although it is generally required for autophagy in plants.
After initiation, the phagophore elongates to cover the peroxisome and become a pexophagosome, which then enters into vacuoles for degradation. Lack of autophagy causes accumulation of damaged peroxisomes and consequently indicates the aggregation of peroxisomes. ATG2 and ATG18a play an indispensable role in enveloping the degraded peroxisomes with ATG8-PE to form pexophagosomes (Fig.  6c). We provided the scheme of process in degradation and formation of the peroxisome aggregation in wild type, peup1/atg2, and peup4/atg7 (Supplementary Fig 19). The difference phenotypes of peroxisome aggregates and dispersion (Fig. 1) may re ect a ATG function in formation of pexophagosome. ATG7 plays a role in the maturation of ATG8-PE as a ubiquitin-activating enzyme-like protein for generating autophagosomes 57,58 . We previously observed ATG8a as dot structures close to degraded peroxisomes in peup1/atg2 and atg5 11,14 . Collectively, these data suggest ATG2, ATG5, ATG7, ATG18a, and ATG8-PE work cooperatively to generate complete pexophagosomes.
We found that the vacuolar membrane surrounded peroxisomes in peup4/atg7 (Fig. 5f-I and Supplementary Fig. 16a), suggesting possible microautophagy during the incorporation of pexophagosomes into the vacuole. Because the process of microautophagy seems incomplete in peup4/atg7, ATG7 and ATG8-PE are probably required in microautophagy. Moreover, in peup4/atg7 cells, bulbs 59 , spherical membrane structures of vacuoles, also interacted with peroxisomes at high frequency (Supplementary Fig. 16 and Supplementary Videos 11,12), suggesting their involvement in microautophagy. Taken together, these ndings suggest that macro-and micro-pexophagy are induced under high-intensity light conditions.
We demonstrated that ATG18a-GFP selectively targets and surrounds peroxisomes to be degraded; this is the rst observation of pexophagosomes forming from phagophores in plant cells. Hence, our analysis gives deep insight into the mechanism of autophagosome formation. Furthermore, our ndings allow further understanding of how plants reduce ROS production via autophagy to improve photosynthetic e ciency and thus increase crop yield.

Plant material and growth condition
Wild-type and transgenic plants were grown in a 16 h light/8 h dark cycle at 23 °C in an incubator (MLR-351, Sanyo Electric Co., Ltd., Japan). Arabidopsis thaliana (L.) Heynh (Columbia, Col-0) and that expressing GFP-PTS1 (the GFP-PTS1 plant) or RFP-PTS1 (the RFP-PTS1 plant) 60 were used as controls.

Imaging analysis
A confocal laser scanning microscope (LSM 510, Zeiss, Germany) with a 40× or 63x objective was used for imaging analyses of peroxisomes and for determining the intracellular distribution of uorescent proteins as described previously 11,36 . The excitation and emission wavelengths for the images were 488 nm and 492-570 nm, respectively, for GFP, and 516 nm and 600-625 nm, respectively, for RFP. Time-lapse images were obtained for 250-300 s with a temporal resolution of 5 s, and movie les were generated with Fiji (ImageJ, NIH public domain). The number of cells and organelles were counted using the Analyze Particles and Cell Counter plugins equipped in Fiji 67 . The size of peroxisome aggregation was measured manually using the polygon selection tool in Fiji after the images were magni ed three-fold for precise selection of the periphery. The pexophagosome around peroxisomes in peup4 targeted by ATG18a-GFP (Fig. 2h) was identi ed by conducting mathematical morphology analysis 28 based on the time-lapse images. Fluorescence intensity (Figs. 2g, 3g, 5e, 6a and Supplementary Fig. 18b) was measured using the Plot Pro le plugin equipped in Fiji. FRAP analysis ( Supplementary Fig. 8) was performed using LSM510 with an Ar laser (488 nm) at 50 % intensity to induce photobleaching. Images were obtained every 1 s, and then uorescence intensity was measured using Fiji.

Measurement of chlorophyll content and photosynthetic e ciency
Chlorophyll content ( Supplementary Fig. 1b) was measured as previously described 68 using the rosette leaves adapted to each light intensity. Photosynthetic e ciency ( Supplementary Fig. 1c) was measured as the maximum yield of photosynthesis system II using a photosynthesis yield analyser (MINI-PAM; Walz, Effeltrich, Germany) 69 using at least three leaves from ve plants after they were adapted to each light intensity. Three independent experiments were performed.
NBT and H 2 -DCF staining Nitro blue tetrazolium (NBT) and 2 7 -dichlorodihydro uorescein (H 2 -DCF) staining were performed as follows: rosette leaves of GFP-PTS1, peup1, and peup4 were immediately submerged in NBT (Sigma-Aldrich) solution for 1 h, and then chlorophyll was repeatedly removed with 100 % ethanol in 95 °C water for 10 min and washed with pure water. In the case of H 2 -DCF staining, the leaves were submerged in 10 µM H 2 -DCF-DA (Thermo Fisher Scienti c) for 10 min and then washed once with pure water 33,34 . At least three independent experiments were performed.

Immunoblot analysis
Immunoblotting was performed as described previously 11 . Total proteins of wild-type, peups, atgs, and various transgenic plants grown under different light intensities for 1-2 d were extracted with extraction buffer containing 10 mM HEPES-KOH (pH 8.0) and a protease inhibitor cocktail (Roche). Then, total proteins were fractionated into supernatant and pellet by centrifugation at 20,000 ´ g for 10 min at 4 °C. The pellets were washed with extraction buffer twice, followed by solubilisation with extraction buffer containing 1% (w/v) SDS. Each 10 µg of total protein was separated by SDS-PAGE and transferred to a polyvinylidene di uoride membrane (Millipore) in a semidry electroblotting system (BioCraft). Then, immunoblot analyses were performed using antibodies against peroxisomal proteins catalase (CAT), peroxin 14 (PEX14), GO, ascorbate peroxidase (APX), and HPR 11 , as well as against mitochondrial proteins cytochrome c oxidase 2 (COXII) (Agrisera, Sweden) and serine hydroxymethyltransferase (SHMT) (Agrisera, Sweden). Signal intensities of bands in the immunoblot image were quanti ed using Dot Blot Analysis in Fiji.
ATG18a-GFP-binding proteins were obtained by immunoprecipitation using µMACS Anti-GFP MicroBeads and µMACS columns (Miltenyi Biotec K.K., USA) 71 . The eluted fraction was assessed by immunoblot analysis using anti-GFP antibody to detect GFP or ATG18a-GFP ( Supplementary Fig. 9a). ATG18a-GFP/GFP-binding proteins were subjected to SDS-PAGE following in-gel digestion. Collected peptides were analysed using nano-LC-MS/MS (LTQ Orbitrap XL; Thermo Fisher Scienti c) 72 . The obtained spectra were searched against the TAIR 10 Arabidopsis protein database (version 20101214) with MASCOT server (version 2.3.02, Matrix Science, London, UK) 73 . The list of identi ed proteins is shown in Supplementary Table 2. The experiments were repeated three times.

Lipid binding assay
The binding ability of ATG18a-GFP to PtdIns3P was determined using PIP Strips P-6001 (Echelon Biosciences Inc., USA) according to the manufacturer's instruction and Tamura et al. (2013) 31 . ATG18a-GFP was isolated as a crude extract from peup1 expressing ATG18a-GFP and incubated with PIP strips for 3 h at 23 °C after removing debris by centrifugation at 1,000 ´ g for 5 min. After two washes with TBS containing 0.1% (v/v) Tween 20, the binding of ATG18a-GFP to lipids was detected using an antibody against GFP and ImageQuant LAS4000 (GE Healthcare) at high sensitivity mode.