AUTOPHAGY-RELATED14 And Its Associated Phosphatidylinositol 3-Kinase Complex Promotes Autophagy In Arabidopsis

Phosphatidylinositol 3-phosphate (PI3P) is an essential membrane signature for both autophagy and endosomal sorting that is synthesized in plants by the class-III phosphatidylinositol 3-kinase (PI3K) complex, consisting of the VPS34 kinase, together with ATG6, VPS15, and either VPS38 or ATG14 as the fourth subunit. While Arabidopsis plants missing the three core subunits are infertile, vps38 mutants are viable but have aberrant leaf, root and seed development, sucrose sensing, and endosomal trafficking, suggesting that VPS38 and ATG14 are non-redundant. Here, we evaluated the role of ATG14 through a collection of CRISPR/Cas9 and T-DNA insertion mutants disrupting the two Arabidopsis paralogs. atg14a atg14b double mutants were relatively normal phenotypically but displayed autophagy defects, including reduced accumulation of autophagic bodies and cargo delivery during nutrient stress. Unexpectedly, homozygous atg14a atg14b vps38 triple mutants were viable but showed severely compromised rosette development and reduced fecundity, pollen germination and autophagy, consistent with a need for both ATG14 and VPS38 to fully actuate PI3P biology. However, the triple mutants still accumulated PI3P, but were hypersensitive to the PI3K inhibitor wortmannin, indicating that the ATG14/VPS38 component is not essential for PI3P synthesis. Collectively, the ATG14/VPS38 mutant collection now permits the study of plants altered in specific aspects of PI3P biology.

Especially prevalent are PPIs phosphorylated at the D-3, D-4, and/or D-5 positions, which are controlled by the combined actions of kinases and phosphatases and then recognized by specific PPI-binding proteins. After docking to PPI-labelled membranes, these binding proteins direct vesicle formation, trafficking and fusion, and ultimately provide signatures for various intracellular compartments. Particularly important in plants is the use of specific PPIs to direct the transport of vesicles from the endoplasmic reticulum (ER) or plasma membrane to the trans-Golgi network for subsequent trafficking via early endosomes and multivesicular bodies to other compartments such as the vacuole (Munnik and Nielsen, 2011;Gerth et al., 2017).
Of interest here are the critical roles of PPIs bearing a phosphate at the D-3 position (PI3P) during endocytosis, endosomal sorting, and macroautophagy (referred to here as autophagy; Noack and Jaillais, 2017; Di Sansebastiano et al., 2018;Marshall and Vierstra, 2018). Despite representing a minor fraction of the PPI pool, PI3P is central to these transport processes through its recognition by a collection of proteins bearing either Fab1/YOTB/Vac1/EEA1 (FYVE) or Phox-Homology (PX) domains that recognize the PI3P moiety (Vermeer et al., 2006;Zhuang et al., 2013;Gao et al., 2015;Noack and Jaillais, 2017). PI3P is particularly important during autophagy, where it works in conjunction with the AUTOPHAGY-RELATED 8 (ATG8) protein modified with the lipid phosphatidylethanolamine (PE) to uniquely decorate the engulfing autophagic membranes as they trap and transport appropriate cargo for final vacuolar breakdown (Marshall and Vierstra, 2018).
In animals, PI3P is synthesized by three types of phosphoinositide 3-kinases (PI3K) containing distinct subunit compositions, defined as class-I, -II, and -III (Balla, 2013). In plants, yeast (Saccharomyces cerevisiae) and algae, only the class-III PI3K/VACUOLAR PROTEIN SORTING 34 (VPS34) type is evident by sequence comparisons. VPS34 specifically phosphorylates phosphatidylinositol at the D-3 position in complex with three accessory proteins: VPS15 and ATG6 (also known as Beclin 1 or VPS30) together with either ATG14 or VPS38 (the ortholog to mammalian UV RADIATION RESISTANCE-ASSOCIATED GENE [UVRAG]) as a fourth subunit, uniquely generating complex-I or complex-II isoforms, respectively (Gerth et al., 2017;Noack and Jaillais, 2017). Three-dimensional structures of the complex-II isoform revealed two membrane contacts, one involving VPS34, which likely positions its kinase catalytic site near inositol head groups, and another involving ATG6 and VPS38/UVRAG, which intertwine through their coiled-coil regions and bind lipids through related Beta-Alpha Repeated Autophagy-Specific (BARA) and BARA2 domains, respectively (Rostislavleva et al., 2015;Young et al., 2016).
ATG14 is devoid of an obvious BARA-type domain but instead has a cysteine-rich region Nterminal to the coiled-coil region that might promote ER association (Matsunaga et al., 2010).
At present, our understandings of PI3K in plants, and the functions of its corresponding lipid product, are limited. Numerous genetic analyses with Arabidopsis (Arabidopsis thaliana) implied that PI3P is essential, based on findings that homozygous null mutants compromising either VPS34, ATG6 or VPS15 could not be generated due to failed reproduction (Fujiki et al., 2007;Qin et al., 2007;Harrison-Lowe and Olsen, 2008;Lee et al., 2008a;Lee et al., 2008b;Patel and Dinesh-Kumar, 2008;Xu et al., 2011;Wang et al., 2012). Whereas female gametogenesis proceeds normally in the mutants and male gametogenesis continues up to formation of the tube cell and the pair of sperm cells, a strong block in pollen germination arises that prevents transmission of the male gametes. Additional defects were also seen in Arabidopsis plants treated with the pharmacological PI3K inhibitors 3-methyladenine, LY294002 and wortmannin, which included compromised autophagy, cellulose biosynthesis, gravitropism, and the vacuolar sorting of seed storage proteins (Takatsuka et al., 2004;Joo et al., 2005;Takáč et al., 2013;Zhuang et al., 2013;Shin et al., 2014;Fujimoto et al., 2015;Huang et al., 2019). VPS34 antisense plants (Welters et al., 1994), and heterozygous atg6 plants or homozygous atg6 plants rescued with low expression levels from an ATG6-GREEN FLUORESCENT PROTEIN (GFP) reporter (Qin et al., 2007;Harrison-Lowe and Olsen, 2008) were also phenotypically aberrant, indicating that the functions of PI3P in plants extend beyond reproduction.
As an alternative to studying the core PI3K complex, we (Liu et al., 2018) and others (Lee et al., 2018;Yuan et al., 2018) have recently focused on the non-redundant VPS38 and ATG14 subunits as a strategy to define the roles of PI3P in plant development. Initial analyses of the predicted VPS38 subunit revealed that it is indeed part of complex-II through its interactions with ATG6 and VPS34. Homozygous null vps38 mutants were viable, indicating that this isoform is not essential during reproduction. However, the mutant plants did exhibit a range of abnormal phenotypes associated with defects in vesicle trafficking and membrane identity, including dampened pollen germination despite apparently normal male gametogenesis, suppressed processing of seed storage proteins and development of storage vacuoles, dilated endosomes, and impaired gravitropism caused in part by aberrant endosomal cycling of the PIN-FORMED (PIN) family of auxin transporters (Lee et al., 2018;Liu et al., 2018;Yuan et al., 2018). Also obvious were defects in leaf morphology and vascular tissue development, and increased seed abortion. Autophagy was also diminished modestly in the vps38 plants, as seen by the reduced accumulation of autophagic bodies inside vacuoles, which in turn elicited a slight hypersensitivity to nitrogen and fixed-carbon starvation reminiscent of weak autophagy mutants (Liu et al., 2018).
To help define the unique functions of ATG14 and its associated VPS34 complex-I kinase, we describe here the genetic, biochemical, and cell biological analysis of the two Arabidopsis paralogs encoding this component. Interaction studies demonstrated that ATG14 likely assembles into a PI3K complex through its interactions with ATG6, and displays expression patterns congruent with those for ATG6, VPS15, and VPS34. Through analysis of atg14a atg14b double null mutants resulting from T-DNA insertions or mutagenesis by clustered regularly interspaced short palindromic repeats (CRISPR) and the CRISPR-associated nuclease Cas9, we found both overlapping and distinct sets of phenotypes as compared to null vps38 mutants, indicating that these subunits of the complex-I and complex-II isotypes have non-redundant roles in plants, like their yeast and mammalian counterparts (Kihara et al., 2001;Itakura et al., 2008).
In particular, homozygous atg14a atg14b plants showed strongly impaired autophagy upon both nitrogen and fixed-carbon starvation, but relatively normal seed and leaf development, and transmission of the mutant alleles.
Furthermore, we successfully generated triple homozygous mutants missing ATG14a, ATG14b and VPS38, strongly suggesting that the fourth subunit in the VPS34 PI3K complex is not essential. The resulting atg14a atg14b vps38 plants displayed a composite phenotype relative to atg14a atg14b and vps38 plants, including a strong suppression of male gamete transmission and autophagy. Surprisingly, the homozygous triple mutants still accumulated PI3P, albeit at reduced levels, but were hypersensitive to wortmannin, strongly suggesting that the three core subunits of the VPS34 PI3K complex are sufficient to synthesize viable levels of PI3P. With this collection of atg14a, atg14b and vps38 alleles, we now provide a venue to study PI3P function(s) throughout the Arabidopsis life cycle.

Identification of Arabidopsis ATG14
As a companion to previous studies on the Arabidopsis VPS34 complex-II PI3K containing its signature subunit VPS38 (Lee et al., 2018;Liu et al., 2018;Yuan et al., 2018), we sought here to investigate the function(s) of complex-I through the analysis of its distinguishing ATG14 subunit.
The two possible paralogs in Arabidopsis (ATG14a and ATG14b) harbored similar domain architectures despite low amino acid sequence identities to each other (46%) and to their yeast counterpart (21% and 34%, respectively;Lee et al., 2018;Liu et al., 2018; Figure 1A). Subsequent searches detected ATG14 relatives in a variety of other plant species (35-100% identity) that were distinct from the VPS38 family and separated phylogenetically into distinct dicot, monocot, and seedless plant clades ( Figure 1A; Supplemental Figure 1A). The ATG14a and ATG14b loci are located on chromosomes 1 and 4, respectively, outside of syntenic patches, suggesting that they originated from a relatively ancient gene duplication event within or before the Brassicaceae lineage.
Unlike plant VPS38, which starts with an N-terminal coiled-coil region (Lee et al., 2018;Liu et al., 2018), the plant ATG14 sequences, like their yeast and mammalian brethren (Diao et al., 2015), contain a cysteine-repeat region upstream of the coiled-coil region ( Figure 1A). This 21-amino-acid stretch is notable for the strict positional conservation of four cysteines (Supplemental Figure 2) that are required for ER localization and autophagy of their yeast and mammalian relatives (Matsunaga et al., 2010). Interestingly, the human Atg14L protein ends in a Barkor Autophagosome-Targeting Sequence (BATS) domain that appears unique to mammals.
Previous studies showed that this BATS region interacts with both PI3P and PI(4,5)P2, possibly through an amphipathic helix, to regulate assembly of the PI3K complex and mediate its association with autophagosomes (Fan et al., 2011;Tan et al., 2016). Whether the plant ATG14 orthologs have a cryptic BATS region with orthologous function(s) remains unclear, but a similar C-terminal helix bearing bulky hydrophobic residues is predicted (Supplemental Figure 2).

ATG14a and ATG14b Have Distinct Expression Patterns
Expression patterns gathered from the Transcriptome Variation Analysis (TraVA) database (http://travadb.org/; Klepikova et al., 2016) revealed that ATG14a and ATG14b, like VPS38, are ubiquitously expressed in most Arabidopsis tissues (Supplemental Figure 1B). In general, transcripts encoding subunits of the VPS34 PI3K complex (except ATG14b) were consistently high in mature or senescing organs, as one might expect for roles of ATG14 and the rest of the complex during cell death and autophagy, as seen in other organisms (Cao and Klionsky, 2007;Lindmo et al., 2008;Jaber et al., 2012;Diao et al., 2015). Similar to mRNAs for the other subunits (ATG6, VPS15, VPS34, and VPS38), ATG14a mRNA was also abundant in anthers (Supplemental Figure 1B), which is consistent with a prominent role for this kinase during pollen development and/or germination (Fujiki et al., 2007;Qin et al., 2007;Harrison-Lowe and Olsen, 2008;Lee et al., 2008b;Xu et al., 2011;Wang et al., 2012), while ATG14b mRNA was abundant in immature and dry seeds, suggesting a specific role for this paralog during reproduction.
Collectively, the expression data indicated that Arabidopsis ATG14a, ATG14b, and VPS38 have both overlapping and distinct functions.
We also examined connections among ATG14, VPS38 and other subunits of the PI3K complex and with other Arabidopsis proteins by co-expression analysis of the Arabidopsis transcriptome, using the ATTED-II version 10 platform (Obayashi et al., 2018). Whereas the VPS34, ATG6, and VPS15 loci displayed direct and indirect expression relationships with a number of other ATG loci, ATG14a displayed a direct co-expression relationship only with ATG18f, while ATG14b and VPS38 displayed direct and indirect co-expression relationships with only ATG13b, respectively (Supplemental Figure 1C), again suggesting non-redundant functions for ATG14 and VPS38. Interestingly, connections with a number of other Arabidopsis proteins, some of which were not yet functionally defined, were also detected that could represent novel autophagic components or regulators.

ATG14 Interacts with ATG6 from the VPS34 PI3K Complex
Given the low sequence homology of Arabidopsis ATG14 relative to its human/yeast counterparts, we attempted to confirm its orthology through complementation of a yeast Datg14 deletion mutant.
Using both the free GFP release assay based on GFP-Atg8 (Marshall et al., 2016), and activation of the Pho8D60 reporter to measure autophagic flux (Noda and Klionsky, 2008), neither full-length Arabidopsis ATG14a nor ATG14b tagged with HA successfully rescued the Datg14 strain, while the yeast ortholog was effective (Supplemental Figure 3). The Arabidopsis counterparts of Vps15, Vps30, and Vps34 (VPS15, ATG6 and VPS34, respectively) also failed to rescue the corresponding yeast deletion strains (Supplemental Figure 3), implying that the plant and fungal VPS34 PI3Ks have sufficiently diverged to prevent cross kingdom complementation. However, we note that Welters et al. (1994) succeeded in rescuing a yeast Dvps34 strain with a chimera of Arabidopsis VPS34 and its yeast counterpart, suggesting that at least some portion(s) of the two proteins are functionally compatible.
As an alternative, we assessed whether ATG14 assembles with the VPS34 PI3K complex through binding studies. As a first attempt, we employed standard Gal4-based yeast two-hybrid (Y2H) assays to test the interaction of ATG14a/b with their likely Arabidopsis partner ATG6, to which they are predicted to bind through their intertwined coiled-coil regions (Rostislavleva et al., 2015;Young et al., 2016). As shown in Figure 1B, clear interactions were detected based on colony growth on selection medium lacking Leu, Trp and His, and containing 3-amino-1,2,4triazole (3-AT), for yeast harboring either ATG14a or ATG14b fused to the Gal4 DNA-binding domain paired with ATG6 fused to the Gal4 activation domain. We note that ATG6 fused to the Gal4 DNA-binding domain self-activated the assay (Lee et al., 2018;Liu et al., 2018). We failed to detect interactions among ATG6, VPS15, and VPS34, or between ATG14a/b and VPS15 or VPS34 using this Y2H system ( Figure 1B).
Next, we tested binding using the split-ubiquitin Y2H system designed for membrane proteins, which exploits the reassociation of ubiquitin fragments to direct the proteolytic release of the membrane-anchored Protein A/LexA/VP16 transcription factor (PLV) (Stagljar et al., 1998).
Notably, this method successfully detected an interaction between ATG6 and VPS34, which was not seen using the Gal4-based system (Liu et al., 2018). Interactions between either ATG14a or ATG14b bearing the C-terminal fragment of ubiquitin fused to PLV (Cub-PLV), and ATG6 bearing the N-terminal fragment of ubiquitin (NubG) were robust on selection medium as compared to the positive control Nub-WT ( Figure 1C). In contrast to VPS38, which bound both VPS34 and VPS15 by this assay (Liu et al., 2018), interactions between ATG14a/b and these two subunits were not evident, suggesting that ATG14 integrates differently into the VPS34 PI3K complex as compared to VPS38.

Isolation of Mutants Compromising ATG14a and ATG14b
To study the function(s) of Arabidopsis ATG14 genetically, we searched the available T-DNA insertion collections for alleles disrupting either locus. Although no insertion lines were available for ATG14a, two lines were described for ATG14b in the Col-0 accession from the Arabidopsis Biological Resource Center (ABRC) collection (www.abrc.osu.edu). One harbored a T-DNA sequence within the 10 th exon at nucleotide 2,155 downstream of the start codon (SAIL_1207_H10) and was designated here as atg14b-1, while the second harbored a T-DNA sequence within the 1 st exon at nucleotide 262 downstream of the start codon (SALK_145203) and was designated here as atg14b-2 (Figures 2A and 2B). We identified homozygous atg14b-1 and atg14b-2 individuals by PCR genotyping. Reverse-transcription (RT)-PCR analyses showed that these mutants fail to accumulate the full-length ATG14b transcript ( Figure 2D), although partial transcripts were evident in atg14b-1, suggesting that shortened polypeptide(s) may be synthesized from this allele. As the hypothetical atg14b-1 protein might encompass most of the ATG14b protein (409 residues), including all of the cysteine-repeat and the coiled-coil regions (Supplemental Figure 2), the atg14b-1 allele might be partially functional.
To generate mutants disrupting ATG14a and additional alleles in ATG14b, we employed CRISPR/Cas9-mediated mutagenesis (Xing et al., 2014;Wang et al., 2015), with guide RNAs that targeted DNA stretches in both ATG14a and ATG14b just upstream of a possible Protospacer Adjacent Motif (PAM) NGG triplet needed to position Cas9 for cleavage (Figures 2A and 2B). This mutagenesis generated three independent alleles for ATG14a, which harbor either 1 bp or 2 bp deletions, or a 1 bp insertion just upstream of the PAM, and were designated here as atg14a-1, atg14a-2, and atg14a-3, respectively (Figure 2A; Supplemental Figure 5A). Importantly, all three mutations introduced a reading frame shift downstream of the Val-20 codon, which should substantially truncate the ATG14a protein if translated.
For ATG14b, we identified three CRISPR/Cas9 alleles that harbor either a 1 bp insertion, a 35 bp deletion combined with a 3 bp insertion, or a 39 bp deletion just upstream of the PAM, and were designated here as atg14b-3, atg14b-4, and atg14b-5, respectively ( Figure 2B). The atg14b-3 allele introduced a frame shift downstream of the Cys-18 codon, which should substantially shorten the translated ATG14b protein ( Figure 2B; Supplemental Figure 5B). The atg14b-4 allele eliminated the junction between the first exon and the first intron resulting in the retention of the first intron. The atg14b-5 allele eliminated 14 codons after the Cys-8 codon, thus deleting the coding sequence for part of the cysteine-repeat region. In all three cases, a substantially truncated, and likely functionally null, ATG14b protein should accumulate if translated (Supplemental Figure 5B).
Within the CRISPR/Cas9 population, we observed a variety of mutant combinations by DNA sequence analysis, including hemizygous individuals and individuals affected at both ATG14a and ATG14b loci. We separated the mutations by selfing and backcrossing to isolate single and double mutant combinations that disrupted both loci but lacked the Cas9 locus. To help track the atg14a-1, atg14a-2, atg14b-4, and atg14b-5 CRISPR/Cas9 alleles, we developed a derived Cleaved Amplified Polymorphic Sequence (dCAPS) strategy based on the loss of the cut sites by the restriction enzymes MseI and RsaI in genomic PCR products derived from the induced alleles. As shown in Figure 2C, only the wild-type ATG14a and ATG14b PCR products were readily digested with MseI and RsaI, respectively. Ultimately, we employed the strong, double-allele combinations atg14a-1 atg14b-4, atg14a-2 atg14b-2, and atg14a-3 atg14b-3 for subsequent studies.
However, like other atg mutants (Doelling et al., 2002;Thompson et al., 2005;Phillips et al., 2008;Chung et al., 2010), the homozygous atg14a atg14b lines displayed a number of phenotypes consistent with a role for ATG14 in autophagy. Indeed, we observed early leaf senescence when grown in short days (SD; 8 h light/16 h dark; Figure 3A), and a hypersensitivity to both nitrogen and fixed-carbon starvation, as evidenced by accelerated chlorosis of seedlings grown on liquid or solid medium in the light but without nitrogen ( Figure 3B; Supplemental Figure   6B), or on sucrose-free medium in extended darkness ( Figure 3C), respectively. Quantification of seedling death caused by fixed-carbon starvation showed that while mutants missing core autophagy components (atg2-1, atg5-1, and atg7-2; Thompson et al., 2005;Chung et al., 2010;Shin et al., 2014) were strongly sensitive and died after 9 d in darkness, the atg14a atg14b double mutants had a weaker response even after 13 d of darkness ( Figure 3D). This mild response was similar to that seen for autophagy mutations compromising the upstream ATG1 kinase complex To ascertain where ATG14a/b influences the autophagy system, we compared levels of various ATG proteins, as well as the ATG12-ATG5 and ATG8-PE adducts. The addition of PE to ATG8 helps embed ATG8 into autophagic membranes, while the conjugation of ATG12 to ATG5 is essential to the ATG12-ATG5-ATG16 ligase complex that drives this lipidation (Marshall and Vierstra, 2018). Levels of ATG8, ATG5, and ATG1 were slightly elevated in the atg14a atg14b double mutants (Supplemental Figure 7A), consistent with prior studies with other atg mutants in which a block in autophagy up-regulates expression of ATG genes and suppresses turnover of the encoded proteins (Contento et al., 2004;Sláviková et al., 2005;Thompson et al., 2005;Phillips et al., 2008;Chung et al., 2010;Li et al., 2014;Huang et al., 2019).
To confirm the identity of the ATG8-PE species, we demonstrated its sensitivity to phospholipase D (Yoshimoto et al., 2004;Chung et al., 2010), which converted the faster migrating lipidated species, seen upon SDS-PAGE in the presence of 6 M urea, to the slower migrating free ATG8 species (Supplemental Figure 7B). Notably, assembly of the ATG12-ATG5 conjugate essential for ATG8 lipidation, and the accumulation of ATG8-PE itself, were not appreciably affected by the absence of ATG14 (Supplemental Figures 7A and 7B). We previously reported that vps38 mutants hyperaccumulate ubiquitin conjugates, possibly due to altered endomembrane dynamics (Liu et al., 2018); we observed no such increase for atg14a-1 atg14b-4 seedlings (Supplemental Figure 7C).

ATG14a/b Promote Autophagic Transport and Accumulation of Autophagic Bodies
Using a GFP-ATG8a reporter (Thompson et al., 2005), we compared the importance of ATG14 and VPS38 in autophagy by fluorescence confocal microscopy to assess autophagic body accumulation, and by the free GFP release assay that measures autophagy-dependent vacuolar transport (Yosimoto et al., 2004;Chung et al., 2010;Suttangkakul et al., 2011;Marshall and Vierstra, 2018). Time-course studies of GFP-ATG8a roots exposed to nitrogen or fixed-carbon starvation and treated with concanamycin A (ConA) to suppress autophagic body degradation (Thompson et al., 2005) revealed that the atg14a atg14b double mutants effectively blocked vesicle accumulation relative to the wild type and the vps38-1 mutant, although not as effectively as the atg7-2 mutant ( Figures 4A and 4B). In fact, under our treatment conditions, the density of autophagic bodies was sometimes so great in wild-type root cells that vesicle clusters appeared ( Figure 4C). Whereas the atg14a atg14b mutants strongly depressed autophagic body accumulation during both nitrogen and fixed-carbon starvation, the vps38-1 mutant was only mildly effective under nitrogen starvation and ineffective during fixed-carbon starvation, demonstrating that ATG14 is uniquely critical for autophagy ( Figure 4B). Measurement of autophagic flux by the free GFP release assays validated the importance of ATG14a/b to autophagy. In contrast to the robust release of free GFP from the GFP-ATG8a reporter in wild-type roots, free GFP was effectively absent in atg7-2 roots under both prolonged nitrogen and fixed-carbon starvation ( Figure 5). The vps38-1 mutant only partially dampened free GFP release upon nitrogen starvation and was relatively ineffective upon fixed-carbon starvation.
By contrast, we saw a strong inhibition of free GFP release in two sets of atg14a atg14b double mutants under both conditions ( Figure 5). Collectively, the data implicate ATG14 and its class-III complex-I kinase as the main source of PI3P during autophagy.

ATG14 is Targeted for Autophagy through its Association with Autophagic Bodies
To confirm the atg14a atg14b phenotypes through complementation and to examine whether ATG14, like other autophagy components, is degraded by autophagy upon binding autophagic membranes (Suttangkakul et al., 2011;Li et al., 2014), we created atg14a-1 atg14b-4 and atg7-2 lines constitutively expressing a translational fusion of ATG14b and GFP. Genomic PCR, RT-PCR, and dCAPs analysis of the resulting homozygous plants detected the full-length GFP-ATG14b coding region with the correct sequence inserted into the respective genomes and transcribed into mRNA (Supplemental Figure 8). Importantly, this reporter successfully replaced ATG14 based on the increased survival of GFP-ATG14b atg14a atg14b seedlings upon fixedcarbon starvation ( Figure 6B). However, we note that the complementation studies were not completely effective based on long-term fixed-carbon starvation assays (Supplemental Figure 9), suggesting either that insufficient levels of GFP-ATG14b accumulated, or that the GFP moiety interfered with ATG14b function.
Analysis of the GFP-ATG14b lines by the free GFP release assay revealed that the GFP-ATG14b reporter was targeted for autophagic turnover. While we detected both fused and free GFP forms in the atg14a-1 atg14b-4 background by immunoblot analysis of total protein extracts and immunoprecipitates generated with anti-GFP antibodies, we only observed the intact fusion in the atg7-2 background ( Figure 6A), consistent with impaired ATG14b turnover. Subsequent confocal fluorescence microscopy of root cells detected autophagic bodies decorated with the GFP-ATG14b reporter. Whereas these fluorescent vacuolar puncta accumulated in nitrogenstarved GFP-ATG14b atg14a-1 atg14b-4 seedlings upon ConA treatment, they were absent in similarly treated GFP-ATG14b atg7-2 seedlings ( Figure 6C). We then confirmed that these puncta were autophagic bodies by co-localization studies on nitrogen-starved atg14a-1 atg14b-4 roots expressing both GFP-ATG14b and mCherry-ATG8a; most vacuolar puncta were labeled with both reporters ( Figure 6D).
We attempted to exploit the GFP-ATG14b fusion to confirm in planta assembly of a PI3K complex also containing VPS34, ATG6 and VPS15 by tandem mass spectrometry (MS) analysis of immunoprecipitates prepared with anti-GFP antibodies from total GFP-ATG14b atg14a-1 atg14b-4 seedling extracts. Unfortunately, we failed to detect any of these factors. Neither were ATG14a/b, VSP34, ATG6, VSP15 or VSP38 detected by shot-gun MS of total seedling extracts despite the identification of over 3,300 distinct Arabidopsis proteins, suggesting that one barrier to detection is that this complex is present at very low levels.

Arabidopsis Plants Missing Both ATG14 and VPS38 are Viable
Given that homozygous Arabidopsis mutants missing the core subunits of the VPS34 PI3K complex could not be created due to defects in pollen development (Fujiki et al., 2007;Qin et al., 2007;Harrison-Lowe and Olsen, 2008;Xu et al., 2011;Wang et al., 2012), we predicted that triple mutants missing ATG14a, ATG14b, and VPS38 function would likewise be embryo-or seedling-lethal. As a first test, we allowed triple heterozygous atg14a-1 atg14b-4 vps38-1 plants to self, and then cataloged the genotypes of the resulting progeny, using genomic PCR to identify the vps38-1 T-DNA insertion and our dCAPS strategy to score the atg14a-1 and atg14b-4 mutations by MseI/RsaI sensitivity (see Figure 2C). After genotyping 523 individuals, we failed to find plants that were homozygous for all three mutations ( Figure 7A). Moreover, while most other allelic combinations segregated at the predicted frequencies, the exceptions were mutants homozygous for two of the loci and heterozygous for the third, strongly suggesting that haploid female and/or male gametes bearing the atg14a atg14b vps38 null combinations were compromised.
As a second test, we screened 96 seedlings derived from a selfed atg14a atg14b vps38 plant that was homozygous for the vps38-1 allele based on the stunted rosette phenotype, and at least heterozygous for the atg14a-1 and atg14b-4 mutations as determined by genotyping.
Remarkably, we identified one individual that was homozygous for all three mutations, strongly indicating that generating Arabidopsis plants null for ATG14a, ATG14b, and VPS38 was possible.
As a third test to improve our odds, we allowed double homozygous/single heterozygous individuals affecting the three loci (Vvaabb, vvAabb, and vvaaBb, where V=VPS38, A=ATG14a, and B=ATG14b) to self, and genotypically examined 48 progeny from each cross, with the calculation that 1 in 4 could be triple homozygous. Strikingly, we succeeded in identifying multiple triple homozygous atg14a atg14b vps38 individuals in each population ( Figures 7B and 7C).
Collectively, their repeated identification demonstrated that the fourth subunit of the VPS34 PI3K is not essential to Arabidopsis.
The actual frequencies for triple homozygous progeny from the third test were significantly lower than expected (6-10%; p-value = 0.0034; chi-squared test), suggesting that one or more aspects of reproduction were compromised but not eliminated. Demonstrably affected was pollen germination (Figures 7D). While atg14a-1 atg14b-4 pollen germinated at high frequency, similar to that seen for wild-type and atg7-2 plants, pollen from homozygous atg14a-1 atg14b-4 vps38-1 plants germinated poorly, with an even lower frequency than that seen for vps38-1 pollen (Liu et al., 2018; Figure 7D; Supplemental Figure 10C). However, triple mutant pollen had normal viability, as judged by Alexander's stain, and underwent normal microsporogenesis, as judged by the presence of the tube nucleus and both sperm nuclei in pollen collected at anthesis (Supplemental Figures 10A and 10B).

Arabidopsis Missing Both ATG14 and VPS38 are Phenotypically Compromised
When grown under non-stressed LD conditions, triple homozygous atg14a atg14b vps38 plants displayed an exaggerated phenotype even when compared to vps38 plants. The rosette leaves were more stunted and crinkled, and flowering was more strongly delayed ( Figures 8A and 8B; Supplemental Figure 11A). The vps38-1 plants flowered ~3 weeks later than the wild type, while the triple mutant plants flowered ~4 weeks later. The triple mutants also senesced faster than vps38-1, were more sensitive to sucrose deprivation, as judged by their slowed root elongation when compared to the atg14a atg14b double and vps38 single mutants, and were more suppressed for hypocotyl elongation when kept in the dark (Supplemental Figure 12). The triple mutants also retained the mild gravitropism insensitivity reported previously for vps38-1 roots (Liu et al., 2018), as evidenced by their enhanced wavy root pattern when grown vertically on plates (Supplemental Figures 12B and 12D).
Surprisingly, seeds from self-crossed atg14a-1 atg14b-4 vps38-1 plants, while not as prevalent, were larger than those from atg14a-1 atg14b-4 and vps38-1 plants, and substantially larger than those from the wild type ( Figure 8C; Supplemental Figure 11B). Imaging of embryonic cotyledons within dry seeds by autofluorescence revealed that the triple mutant contained cells more tightly packed with protein storage vacuoles (PSVs; Figure 8D). When seeds from the triple mutant combinations were subjected to SDS-PAGE and either stained for total protein or subjected to immunoblot analysis with antibodies against the 12S globulin and 2S albumin seed storage proteins ( Figure 8E; Supplemental Figure 11C), we determined that unprocessed storage protein precursors accumulated, indicative of defects in membrane trafficking to PSVs, as seen previously for the vps38 mutants (Liu et al., 2018). Accordingly, the level of VACUOLAR SORTING RECEPTOR 1 (VSR1) critical for depositing storage proteins into PSVs (Shimada et al., 2003;Di Sansebastiano et al., 2018) was also strongly elevated in atg14a-1 atg14b-4 vps38-1 seeds, moderately increased in vps38-1 seeds, but barely detected in wild-type, atg14a-1 atg14b-4, or atg7-2 seeds ( Figure 8E). Taken together, the phenotypes of the triple mutant appear additive to those missing ATG14 and VPS38 individually.

The atg14a atg14b vps38 Triple Mutant has Strongly Dampened Autophagy
To assess how the lack of both ATG14 and VPS38 would impact autophagy, we compared various aspects of the triple mutant relative to the atg14a atg14b and vps38 mutant combinations.
By contrast, the ATG8-PE adduct formed and accumulated normally in the triple mutant ( Figure   9B), indicating that lipidation was unaffected.
When measuring autophagic flux with the GFP-ATG8a reporter, either by the free GFP release assay or by confocal fluorescence microscopy, the triple mutant appeared more compromised relative to the atg14a atg14b and vps38 backgrounds (Figures 9C-E). This suppression was seen by the dampened accumulation of free GFP upon either nitrogen or fixed-carbon starvation, and by the reduced accumulation of autophagic bodies relative to those seen in the wild type. Comparisons of the mutant panel indicated that, while ATG14 played a dominant role in autophagy, VPS38 also had an effect. However, the triple mutant did not compromise autophagy as much as in the atg7-2 background ( Figure 9E), suggesting that autophagy can proceed slowly without the fourth subunit of the VPS34 PI3K complex.

Loss of ATG14 and VPS38 Suppresses PI3P Accumulation
We hypothesized that the phenotypes associated with the atg14a atg14b and/or vps38 mutants were generated by attenuated PPI synthesis. While prior studies with the vps38-1 mutant reported no effect on PPI levels generally, or PI3P levels specifically (Lee et al., 2018), it was unclear how atg14a atg14b double mutants or the triple mutant would respond. Here, we labeled wild-type and mutant seedlings with [ 32 P]-PO4 3-, and then analyzed their PPI content by thin layer chromatography (TLC). As shown in Figure 10A, we saw little to no change in the profiles of various phospholipids in atg14a atg14b vps38 seedlings versus the wild type, implying that the ATG14/VPS38 subunits of the VPS34 PI3K have little influence over phospholipid metabolism overall.
To specifically measure the levels of PI3P and its relatives PI4P and PI(4,5)P2, we excised the PPI spots after TLC, removed their fatty acids by mono-methylamine deacylation, and separated the resulting glycerophosphoinositol phosphates (GroPIPs) by anion-exchange high performance liquid chromatography (HPLC). We then quantified the levels of individual species by scintillation counting of the HPLC fractions for 32 P (Munnik, 2013). The vps38-1 mutant only modestly reduced the levels of PI3P, with a slightly greater reduction evident for atg14a atg14b lines ( Figure 10B). Surprisingly, when we examined the atg14a atg14b vps38 triple mutant, the accumulation of PI3P was still evident but was now dampened approximately 50% compared to that seen in the wild type ( Figure 10B). In fact, all four different triple mutant seed batches derived from the double homozygous single heterozygous parents described in Figure 7B (Vvaabb, vvAabb and vvaaBb) accumulated significantly less PI3P than the atg14a atg14b double or the vps38 single mutants. We also observed similar but more modest reductions in PI4P levels in the triple mutant, while on the contrary, PI(4,5)P2 levels increased in all backgrounds that harbored the vps38-1 mutation ( Figure 10B; Supplemental Figure 13), indicating that alterations in VPS34 PI3K activity perturbed the levels of related PPIs. Notably, parallel analysis of PPI levels in atg7-2 seedlings revealed that a general block in autophagy also modestly altered the pools of both PI3P, PI4P, and PI(4,5)P2 ( Figure 10B). Taken together, the atg14a atg14b vps38 triple mutations reduced but did not eliminate PI3P.

vps38 Mutants are Hypersensitive to Wortmannin
We speculated that the residual PI3P seen in the triple mutants was derived from the core VPS34 PI3K complex working even in the absence of ATG14 and VPS38. To indirectly test this hypothesis, we examined the sensitivity of 10-d-old wild-type seedlings, and various allelic combinations of atg14a-1, atg14b-4 and vps38-1, to wortmannin, a well-described inhibitor of VPS34 (Takatsuka et al., 2004: Huang et al., 2019. Whereas growth of wild-type and atg14a-1 atg14b-4 seedlings was relatively immune to concentrations of wortmannin sufficient to suppress autophagy (Shin et al., 2014), growth of the vps38-1 and the triple atg14a-1 atg14b-4 vsp38-1 seedlings was severely compromised at concentrations of 5 µM and above ( Figures 10C and   10D). Collectively, these results suggested that the viability of vps38 mutants is derived from the VPS34/ATG6VPS15 core complex working alone to synthesize PI3P, and indicated that PI3P associated with endosomal trafficking is more phenotypically critical to Arabidopsis growth and development than that associated with autophagy.

DISCUSSION
Defining the roles of PI3P in plant endosomal trafficking and autophagy have been challenging due to the lack of viable mutants missing the core components of the VPS34 PI3K complex responsible for synthesizing this phospholipid (Fujiki et al., 2007;Qin et al., 2007 Gao et al., 2015). However, given the inherent secondary effects often associated with these inhibitors, the results presented here with ATG14 now directly connect this phosphoinositide to autophagy. Although sequence homology between Arabidopsis ATG14a/b (and other plant versions) and their animal and yeast counterparts is low (< 20%), their identities were confirmed by Y2H and BiFC interaction assays with ATG6. Such low identities also exist among the three core subunits of the class-III complex, and were likely responsible for our failure to rescue any of the yeast mutants (Δatg14, Δvps15, Δvps30 (Δatg6), and Δvps34) with their Arabidopsis counterparts. We note that VPS34 and VPS15 failed to associate with each other or with ATG6 in our Gal4-based Y2H binding assays, but were previously shown to associate by split-ubiquitin Y2H and/or BiFC assays (Liu et al., 2018), indicating that the interactions within the PI3K complex are highly sensitive to the binding assay chosen. phenotypically, except for strongly attenuated autophagy and the associated hypersensitivity to nitrogen and fixed-carbon starvation and premature rosette senescence. Whereas the collection of atg14a atg14b mutants effectively suppressed delivery of autophagic cargo to vacuoles (as seen by the free GFP release assay and confocal microscopy using the GFP-ATG8a reporter), in agreement with this isoform being the primary effector of such recycling, the process was also more subtly compromised in vps38 mutants, which was especially noticeable during nitrogen starvation. The reason behind this response to nitrogen starvation is unknown, but may reflect additional roles for VPS38 and PI3P in plants under nitrogen stress.
From the analysis of the GFP-ATG14b reporter, it appears that ATG14 is also an autophagy substrate, similar to other regulatory components of autophagy such as the ATG1, ATG11, and ATG13 subunits of the ATG1 kinase complex (Suttangkakul et al., 2011;Li et al., 2014). Whereas the reporter was stable when expressed in the atg7-2 background (as judged by the free GFP release assay), its GFP moiety was released in the atg14a-1 atg14b-4 background, consistent with its delivery to vacuoles by the autophagic system. The fact that this release was seen in non-stressed plants indicates that GFP-ATG14b constitutively undergoes autophagy. This delivery was confirmed by confocal fluorescence microscopy, which showed the appearance of GFP fluorescence within vacuolar puncta decorated with mCherry-ATG8a by a process also requiring ATG7. It should be noted that expression of GFP alone in Arabidopsis, or of a truncated des-Gly version of GFP-ATG8 blocked in lipidation, does not lead to fluorescence visualization of autophagic bodies in vacuoles (Yoshimoto et al., 2004;Thompson et al., 2005;Marshall et al., 2015), thus discounting the possibility that the freed GFP became an autophagy substrate after cytoplasmic cleavage.
Presumably, the autophagic breakdown of the GFP-ATG14b reporter (and possibly the rest of the VPS34 PI3K complex), is driven by its association with developing autophagic membranes, which might trap the complex before enclosure. In fact, all three core subunits of the class-III kinase complex in mammals have likely ATG8-interacting motifs (AIMs) (Birgisdottir et al., 2019) that could indirectly tether ATG14 to autophagic membranes before enclosure, while the cysteine-repeat and BATS domains in ATG14 might directly promote membrane association (Fan et al., 2011).
To test whether simultaneous loss of both ATG14 and VPS38 would phenocopy loss of the essential core subunits of the class-III PI3K complex (ATG6, VPS15, and VPS34) (Fujiki et al., 2007;Qin et al., 2007;Harrison-Lowe and Olsen, 2008;Xu et al., 2011;Wang et al., 2012), we attempted to generate triple homozygous atg14a atg14b vps38 plants. While initial screens of progeny generated by selfing of triple heterozygous plants failed, despite the analysis of a population predicted to be sufficient (> 500 progeny), we eventually succeeded by using various combinations of double homozygous/single heterozygous parents in self crosses. Even with these populations, our success in finding triple homozygous progeny was well below expectations. We presume that this low success rate was at least partially explained by the severely attenuated germination of atg14a atg14b vps38 pollen, mainly caused by the absence of VPS38 impacting PI3P levels (Liu et al., 2018;this report). However, we note that PI(3,5)P2, which is synthesized from PI3P through the FATTY ACID BIOSYNTHESIS 1 (FAB1) and FAB2 and VACUOLE MORPHOLOGY AND INHERITANCE MUTANT 14 (VAC14) factors, has also been connected to proper pollen development (Whitley et al., 2009;Zhang et al., 2018). Although the consequences of fab1/2 and vac14 mutants on pollen appear distinct from those seen here for atg14a atg14b vps38 mutants, they do raise the possibility that the defects in pollen germination seen here actually reflect diminished PI(3,5)P2 levels.
Owing to the challenges in isolating null mutants in the core PI3K subunits, our analyses of homozygous atg14a atg14b vps38 Arabidopsis provide a compelling genetic demonstration of PI3P importance in plants. Notably, triple homozygous plants had a collection of abnormal phenotypes that appeared to combine the defects seen for atg14a atg14b and vps38 lines separately, again consistent with the conclusion that the two versions of the VPS34 PI3K complex mainly work non-redundantly. However, some overlap in functions was apparent by the increased effects of the triple mutant on phenotypes associated with either ATG14 or VPS38. These included accentuated defects in leaf morphology, pollen germination, seed size, and seed storage protein processing, which appear associated with VPS38, and accentuated defects in autophagy, which appear associated with ATG14. Interestingly, neither the atg14a atg14b mutants nor the triple mutant disturbed assembly of the ATG12-ATG5 and ATG8-PE adducts, suggesting that PI3P acts somewhere after decoration of the phagophore with ATG8-PE but before fusion of autophagosomes with the vacuole. Wherever PI3P acts, the suppressed autophagy seen for atg14a atg14b backgrounds confirms that autophagy vesicle dynamics requires both ATG8-PE and PI3P, presumably through autophagic adaptors that recognize both membrane surface signatures (e.g., the SH3 domain-containing protein 2 (SH3P2); Zhuang et al., 2013).
From quantitative analysis of various PPIs, we were surprised to see that the levels of total PPIs generally, and PI3P specifically, were not strongly compromised in the homozygous atg14a atg14b vps38 plants. At most, we saw only a ~50% reduction in PI3P upon analysis of four different seed lines, along with a modest drop in PI4P and a slight increase in PI(4,5)P 2 levels.
The most parsimonious explanation is that ATG14 and VPS38 promote, but are not essential for, the activity of the VPS34 PI3K, possibly by helping tether the complex to appropriate membrane surfaces. The residual PI3P levels seen in atg14a atg14b vps38 plants may then reflect the ATG6/VPS15/VPS34 core acting alone but in a less directed fashion. The phenotypic hypersensitivity of the vps38 backgrounds to low concentrations of the VPS34-specific inhibitor wortmannin supports this notion, which we presume reflects suppression of residual VPS34 activity focused toward endosomal trafficking. That growth of the atg14a atg14b mutants was not appreciably compromised by wortmannin was consistent with the relatively normal phenotype of Arabidopsis autophagy-null mutants under non-stressed conditions (Doelling et al., 2002: Thompson et al., 2005Supplemental Figure 6). However, we cannot discount the remote possibility that plants either employ a novel PI3K type that is unrelated to complexes-I and -II typical of eukaryotic class-III PI3Ks, or have additional regulators for the VPS34 PI3K complex beyond ATG14/VPS38 (e.g., mammalian Rubicon, for which a possible Arabidopsis homolog exists (Matsunaga et al., 2009;Heucken and Ivanov, 2018)).
Why PI(4,5)P 2 levels increased in the vps38 and atg14a atg14b vps38 backgrounds is currently unclear. In two simple scenarios, this increase might have been induced by a delay in Arabidopsis seeding development in the absence of PI3P, or might represent an attempt by the mutants to compensate for reduced PI3P levels in an effort to maintain normal endomembrane dynamics and/or autophagy. Of interest to the latter scenario, Tan et al. (2016) recently connected PI(4,5)P2 to autophagy in mammalian cells through the ability of the ATG14 BATS domain to bind this PPI, which in turn promoted class-III PI3K assembly. Alternatively, increased PI(4,5)P2 levels may reflect the accumulation of intermediates of membrane trafficking decorated with this lipid that are unable to reach their final destinations in the absence of PI3P. In any event, our assembly of a viable suite of Arabidopsis mutants missing various combinations of ATG14a, ATG14b, and VPS38 now permits the study of PI3P across the entire life span of Arabidopsis. When combined with various endomembrane markers and VPS34-specific PI3K inhibitors such as wortmannin, 3methyladenine and LY249002 (Takatsuka et al., 2004;Takáč et al., 2013;Zhuang et al., 2013;Shin et al., 2014;Huang et al., 2019), understanding the precise roles of this PPI in endomembrane dynamics and autophagy should now be possible.

Sequence and Phylogenetic Analyses of ATG14
We type "Cytoscape", Coex option "Add many genes", and PPI option "Draw 2-path PPIs". For RT-PCR analysis, Total RNA was extracted using the plant RNeasy mini kit (QIAGEN) from 14-d-old seedlings grown on Murashige and Skoog (MS) medium. mRNAs were converted to cDNA via the SuperScript III first-strand synthesis system (Invitrogen) with oligo(dT)20 primers, before using them as templates for RT-PCR with gene-specific primers. Details of all oligonucleotide primers used are given in Supplemental Data Set 1.

Protein-Protein Interaction Assays
Standard Gal4-based Y2H assays relied on HIS3 expression, as described (Gingerich et al., 2007), with slight modifications. cDNAs encoding the full-length sequence of ATG14a, ATG14b, ATG6/VPS30, VPS15, and VPS334 were generated by PCR with the Fusion DNA polymerase (Thermo Fischer Scientific) using reverse-transcribed total seedling RNA isolated from wild type Col-0, before subcloning into the pDONR221 entry plasmid via Gateway BP clonase II reactions (Invitrogen). These fragments were recombined in-frame into the pDEST22 plasmid harboring the Gal4 activation domain (AD) and the pDEST32 plasmid harboring the Gal4 DNA-binding domain (BD) via Gateway LR clonase II reactions (Invitrogen). The plasmids were co-transformed pairwise into yeast strain MaV203, and the colonies were assayed for positive protein-protein interactions by 2-d growth at 30ºC on synthetic dropout medium lacking His, Leu and Trp, and containing 25 mM 3-AT, using synthetic dropout medium lacking Leu and Trp as the growth control, and unrecombined pDEST22 and pDEST32 as negative interaction controls.
For Y2H assays using the split-ubiquitin Protein A/LexA/VP16 (PLV) system, positive interactions are based on association between the N-terminal and C-terminal fragments of ubiquitin (NubG and Cub, respectively), which allows proteolytic release and nuclear import of the chimeric PLV transcription factor to then activate HIS3 and ADE2 expression (Stagljar et al., 1998;Liu et al., 2018). We introduced the relevant cDNAs into the pMETYCgate/pNXgate32/pXNgate22-based plasmid system by Gateway recombination as above. Candidate baits were translationally fused to the N-terminus of Cub-PLV and transformed into yeast strain THY.AP4 (mating type a), while candidate preys were fused to either the Nterminus or C-terminus of NubG and transformed into yeast strain THY.AP5 (mating type α). After mating, the diploid THY.AP4 THY.AP5 cells were plated on synthetic dropout medium lacking His, Leu, Trp, Ura, and Ade to test for protein-protein interactions by growth. The NubG and Nub-WT empty vectors were used as negative and positive controls, respectively (Stagljar et al., 1998).

Yeast Complementation by Free GFP Release and Pho8Δ60 Assays
We PCR-amplified the coding sequences for the Arabidopsis or yeast ATG14, VPS15, ATG6/VPS30, and VPS34 genes from cDNAs generated from total RNA extracted from Col-0 seedlings or BY4741 yeast cells using the SuperScript III first-strand synthesis system as above.
Where necessary, sequences for an N-terminal HA tag in the relevant PCR amplification primer were included. The resulting PCR products were recombined into pDONR221 via Gateway BP clonase II reactions as above, sequence-confirmed, and then recombined into the pAG424-GPD- The Pho8Δ60 activity assays measuring autophagic flux were performed in yeast essentially as described (Noda and Klionsky, 2008), with minor modifications (Marshall et al., 2016). We grew strains TN124 or RSM353-365 as 250 mL YPDA cultures, subjected them to nitrogen starvation for 8 h, and collected aliquots corresponding to 5.0 OD units at the indicated time points. Cell pellets were resuspended in 500 µL lysis buffer (20 mM piperazine-N,Nʹ-bis-2ethanesulfonic acid-KOH pH 8.5, 50 mM KCl, 100 mM potassium acetate, 10 mM MgSO4, 10 μM ZnSO4, 0.5% (v/v) Triton X-100, and 1 mM phenylmethylsulphonyl fluoride), lysed by vigorous mixing in the presence of acid-washed glass beads, and clarified by centrifugation at 20,000g for 20 min at 4ºC. Equal amounts of protein (20 μg) were then assayed spectrophotometrically for alkaline phosphatase activity using p-nitrophenyl phosphate as the substrate.

Plant Materials and Growth Conditions
The atg14b-1 (SAIL_1207_H10) and atg14b-2 (SALK_145203) T-DNA insertion mutants in the Arabidopsis Col-0 accession were obtained from the ABRC. We generated atg14a and additional atg14b mutants with an egg cell-specific promoter-controlled CRISPR/Cas9 and RsaI restriction sites in the atg14a and atg14b PCR products, respectively, as assayed by agarose gel electrophoresis of digested PCR fragments.
For atg14 complementation studies, we constructed a GFP-ATG14b transgene under the control of the 35S promoter from a cDNA encompassing the full-length coding region, which was introduced into the pMDC43 plasmid (Curtis and Grossniklaus, 2003) between the 35S promoter and the NOS 3′ UTR sequences using Gateway LR clonase II reactions, resulting in the generation of the 35Spro:GFP-ATG14b construct, which was introduced into the homozygous atg14a-1 atg14b-4 and atg7-2 backgrounds by the floral-dip method, followed by selfing to obtain fully homozygous lines.

Plant Phenotypic Assays
We surface-sterilized seeds with the vapor-phase method before sowing seeds in culture dishes containing full-strength MS medium supplemented with 2% sucrose, 1.4% agar, and 2 mM 2-(Nmorpholino)-ethanesulfonic acid (MES)-KOH pH 5.7. Seeds were stratified for 3-4 d, and then To assay wortmannin sensitivity, we stratified seeds at 4ºC for 5 d before germination on solid GM medium (3.2 g/l Gamborg's B5 basal medium with minimal organics, 1% (w/v) sucrose, 0.05% (w/v) MES-KOH (pH 5.7), and 0.7% (w/v) agar) supplemented with the indicated concentration of wortmannin (Sigma-Aldrich catalog number W1628) or equivalent volumes of DMSO as the control. After 10 d of growth at 21 to 23ºC under a LD photoperiod, we measured the fresh weight of 10 seedlings.

Immunoblot and Immunoprecipitation Assays
We detected ATG1, ATG8, ATG5, the ATG12-ATG5 conjugate, GFP-ATG8a, and ubiquitin conjugates by immunoblot assays using total seedling extracts prepared from 100 mg of seedlings homogenized in 200 µL of 2X SDS-PAGE sample buffer (100 mM Tris-HCl (pH 6.8), 4% SDS, 0.2% bromophenol blue, and 20% glycerol), followed by heating the samples to 95ºC for 2 min, and then clarifying the extracts by brief centrifugation at 14,000g for 10 min at room temperature.
ATG8a lipidation was monitored according to Chung et al. (2010). Membrane fractions were collected from total seedling extracts by centrifugation at 100,000g for 10 min at 4ºC. The pellets were solubilized in 0.5% (v/v) Triton X-100 in TNPI buffer (50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride, and 10 mM iodoacetamide) and clarified, and the resulting supernatants were incubated for 1 h at 37ºC with or without the addition of Streptomyces chromofuscus phospholipase D (Enzo Life Sciences product number BML-SE301). Samples were subjected to SDS-PAGE electrophoresis in the presence of 6 M urea and immunoblotted with anti-ATG8a antibodies. For the GFP-ATG8a cleavage assays (Chung et al., 2010;Suttangkakul et al., 2011;Huang et al., 2019), we grew seedlings expressing GFP-ATG8a in continuous white light in nitrogen-containing liquid MS medium before transfer to MS medium lacking nitrogen, or to MS medium lacking sucrose and placed in darkness. Immunoblots were performed on total seedling extracts with anti-GFP antibodies as above.
For the immunoprecipitation of GFP-ATG14b, we grew seedlings of the indicated genotype (Col-0 [WT], GFP-ATG14b atg14a-1 atg14b-4 or GFP-ATG14b atg7-2) for 7 d in 50 mL liquid GM medium at 21ºC to 23ºC in constant white light with gentle shaking, with ∼100 mg of dry seeds used per culture, resulting in ∼5 g of fresh weight tissue. Frozen tissue was ground to a fine powder in liquid nitrogen, and proteins were then extracted on ice for 20 min with 1.5 volumes of extraction buffer (50 mM HEPES (pH 7.5), 150 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 5 mM DTT, 2 mM PMSF, 0.1% (v/v) Triton X-100, and 1X plant protease inhibitor cocktail). Extracts were filtered through two layers of Miracloth, clarified at 30,000g for 20 min at 4ºC, and the resulting supernatants immediately applied three times at 4ºC over a 12 mL PolyPrep chromatography column containing 100 μL (equal to a 50 μL bead volume) of GFP-Trap_A beads (ChromoTek product number gta-20) pre-equilibrated in extraction buffer. The column was washed five times with wash buffer (50 mM HEPES (pH 7.5), 150 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, and 2 mM DTT), the remaining bound proteins were eluted with 100 μL of 200 mM glycine-HCl (pH 2.5), and immediately neutralized with 20 μL of 1 M Tris-HCl, pH 8.0). The input and elution samples were analyzed by SDS-PAGE followed by immunoblotting with anti-GFP or anti-histone H3 antibodies. The input sample was concentrated 200-fold by precipitation with 50% TCA prior to analysis. The GFP-ATG14b immunoprecipitates and total cell lysates were trysinized prior to HPLC-coupled tandem MS analysis as described by Marshall et al. (2019).
Seed protein extraction and immunoblot assays were performed as previously described (Shimada et al., 2003;Reguera et al., 2015). Briefly, we homogenized dry seeds (60 each) in 150 µL of 2X SDS-PAGE sample buffer, clarified by centrifugation at 14,000g for 10 min at 4ºC, subjected the supernatants to SDS-PAGE, and either stained for protein with Coomassie Brilliant Blue R-250 or immunoblotted with antibodies against either Arabidopsis 2S albumin, 12S globulin α-subunit, or VSR1 (Reguera et al., 2015).

Confocal Fluorescence Microscopy and Image Analysis
Fluorescent cell images were collected with a Nikon A1Si laser scanning confocal microscope, using 358-nm light combined with a 461-nm filter for DAPI, 488-nm light combined with 500-550nm filters for GFP, YFP, and seed autofluorescence, and 543-nm light combined with a 565-615nm filter for mCherry. The images were processed using the NIS Elements software (Nikon) and converted to images TIFF files for use in the Figures. We quantified the accumulation of autophagic bodies using ImageJ (www.imagej.nih.gov/ij) by counting their numbers in 100 µm 2 areas of the central vacuole of multiple cells from multiple roots (25 cells total) after 4, 8 or 16 h of nutrient starvation.
The next day, lipids were extracted and separated by thin layer chromatography (TLC) using an alkaline solvent, and the TLC plates were exposed to X-ray film for visualization and quantification by phosphoimaging (Typhoon FLA 7000; GE Healthcare).
To distinguish PI3P from PI4P, we scraped the phosphatidylinositol monophosphate (PIP) spots from the TLC plates and subjected them to deacylatation by mono-methylamine (Sigma-Aldrich product number 426466) as described (Munnik, 2013). The resulting GroPIPs were desalted through cation exchange (AG 50W-X8, 200-400 mesh; Bio-Rad) and separated them by HPLC using a Partisil 10 strong anion-exchange column (4.5 x 250 mm; Sigma Aldrich) and a gradient of NaH2PO4 (pH 3.7). Fractions (collected every 20 sec) were quantified for radioactivity by liquid scintillation counting. The final results for different PIP isomers were obtained by quantifying the PIP spots from the TLC by phosphoimaging and then normalizing the results with respect to the ratios of PI3P to PI4P obtained by HPLC analysis.

Statistical Analysis
Data reported in this study are means (±SD) of at least three independent experiments, each consisting of three technical replicates, unless otherwise noted. p-values were determined by one-way analysis of variance (ANOVA) or two-tailed Student's t-tests (Supplemental Data Set 3).
The outcomes of the genetic crosses were evaluated by the chi-squared test (p-value < 0.05).

Accession Numbers
DNA and derived protein sequence data from this article are available in the Arabidopsis Information Resource database under the following accession numbers: ATG14a (At1g77890),