COST1 regulates autophagy to control plant drought tolerance

Significance

Drought is one of the most devastating threats to agricultural sustainability worldwide. Autophagy is known to be critical for plant responses to multiple stresses, including drought, but a direct link between drought tolerance and autophagy is still lacking. We report here the identification of a plant-specific protein, COST1 (constitutively stressed 1), which attenuates autophagy under optimal growth conditions, thus controlling the trade-off between plant growth and stress tolerance. In addition to expanding our understanding of the regulation of autophagy in plants, the enhanced drought tolerance of a cost1 mutant and the high conservation of COST proteins throughout the plant kingdom also indicates a potential for engineering COST1 to enhance stress tolerance in crops.

Abstract

Plants balance their competing requirements for growth and stress tolerance via a sophisticated regulatory circuitry that controls responses to the external environments. We have identified a plant-specific gene, COST1 (constitutively stressed 1), that is required for normal plant growth but negatively regulates drought resistance by influencing the autophagy pathway. An Arabidopsis thaliana cost1 mutant has decreased growth and increased drought tolerance, together with constitutive autophagy and increased expression of drought-response genes, while overexpression of COST1 confers drought hypersensitivity and reduced autophagy. The COST1 protein is degraded upon plant dehydration, and this degradation is reduced upon treatment with inhibitors of the 26S proteasome or autophagy pathways. The drought resistance of a cost1 mutant is dependent on an active autophagy pathway, but independent of other known drought signaling pathways, indicating that COST1 acts through regulation of autophagy. In addition, COST1 colocalizes to autophagosomes with the autophagosome marker ATG8e and the autophagy adaptor NBR1, and affects the level of ATG8e protein through physical interaction with ATG8e, indicating a pivotal role in direct regulation of autophagy. We propose a model in which COST1 represses autophagy under optimal conditions, thus allowing plant growth. Under drought, COST1 is degraded, enabling activation of autophagy and suppression of growth to enhance drought tolerance. Our research places COST1 as an important regulator controlling the balance between growth and stress responses via the direct regulation of autophagy.

Drought is an increasing problem worldwide because of water shortage and climate change, and is responsible for substantial yield loss in crops (1, 2). Plants have evolved sophisticated mechanisms for sensing and responding to drought, which involve perception of the stress, signal transduction, and physiological adaptions (3–5). Major signaling pathways have been identified for plant responses to drought, including those triggered by the drought-inducible phytohormone abscisic acid (ABA) (3) and the reactive oxygen species H₂O₂. Drought stimulates the biosynthesis of ABA (6, 7), which is subsequently bound by its receptor PYR/PRL/RCAR, leading to activation of the SnRK2 (SNF1-related kinases subfamily 2) family protein kinases (8–11). Active SnRK2 kinase then phosphorylates SLAC1 (S-type anion channel) and KAT1 (potassium channel), thus modulating stomatal movement (12–14).

Two plasma membrane NADPH (nicotinamide adenine dinucleotide phosphate, reduced form) oxidases, respiratory burst oxidase homologs D and F (RbohD/F), are responsible for the production of H₂O₂ (15), which can activate Ca²⁺ channels to favor stomatal closure (16). RbohF is phosphorylated by OST1 (open stomata 1) to favor H₂O₂ production (17). A receptor-like kinase family protein GHR1 (guard cell hydrogen peroxide-resistant 1) is a positive regulator of drought responses that acts downstream of H₂O₂ signaling (18). Genetic studies indicate an interplay between ABA and H₂O_2, suggesting that ABI1 acts upstream of H₂O₂ while ABI2 is downstream (19).

The 26S proteasome-mediated protein turnover plays a critical role in plant stress responses, including those to drought (5, 20). During stress conditions, targeted proteins are specifically recognized and ubiquitinated via a sequential relay of an E1 activation enzyme, E2 conjugation enzyme, and an E3 ligase (21). Poly-ubiquitinated proteins are then subjected to degradation by the 26S proteasome, providing a mechanism for precise regulation of signaling (21). Protein turnover during various environmental stresses can also be mediated by autophagy (22, 23). Upon the induction of autophagy, a double-membrane structure engulfs unwanted cellular components, including individual proteins, protein aggregates, and organelles, forming an autophagosome, and transports them to the central vacuole for degradation and recycling (24). A number of genes, termed ATG (autophagy-related) genes, have been identified as required for autophagosome formation or delivery to the vacuole (25, 26). In addition, specific cargo can be selected for autophagy by cargo receptors, such as NBR1 (neighbor of BRCA1), which recognizes protein aggregates (27–29). Autophagy has a pivotal role in plant responses to a broad array of biotic and abiotic stresses, including drought, salt, oxidative stress, flooding, and pathogens (30–34).

In this study, we identified a plant-specific DUF641 family protein that we named COST1 (constitutively stressed 1). A cost1 mutant has defects in development, a strong drought-resistance phenotype, and constitutive activation of autophagy. Consistently, overexpression of COST1 leads to drought hypersensitivity and decreased autophagy in response to stress. The drought tolerance of the cost1 mutant requires an intact autophagy pathway and COST1 interacts directly with ATG8e, a key factor required for autophagy. We propose a model in which COST1 negatively regulates autophagy, and aggregation/degradation of COST1 during drought therefore activates autophagy, allowing plants to tolerate drought conditions.

Results

A cost1 Mutant Has Pleiotropic Growth Defects.

To date, in the genome of Arabidopsis thaliana, over 40% of all protein-encoding genes are classified as having unknown function or are poorly characterized (35), despite many efforts to identify gene functions (36, 37). Here, we identified a plant-specific protein (AT2G45260) containing a DUF641 (domain of unknown function 641) domain. We named this gene COST1, based on the stress-tolerant phenotype of a cost1 knockout mutant we later observed (see below). BLAST search against TAIR10 (The Arabidopsis Information Resource, arabidopsis.org) and alignment analysis indicated that 13 COST-related proteins are present in A. thaliana and share sequence similarity at their N termini, which contains the DUF641 domain (Fig. 1A and SI Appendix, Fig. S1). Based on precomputed results from Pfam (38) and sequence analysis (Materials and Methods), DUF641 family proteins (referred to as COST proteins) are plant-specific and can be found widely throughout land plants, including mosses, but are absent from algae (Fig. 1B and SI Appendix, Fig. S2). COST proteins can be classified into six subfamilies, and the duplication between the subfamily containing COST1 and subfamily Branch_1 happened before the divergence of seed plants (Fig. 1B and SI Appendix, Figs. S1 and S2).

Fig. 1.

A conserved plant-specific COST1 (ID: AT2G45260) protein required for normal growth and development. (A) Schematic diagram of the structure of the COST1 protein; the conserved DUF641 domain is indicated at the N terminus (N); numbers below indicate the position of amino acids. (B) Phylogenetic analysis of COST proteins in different plant species. The numbers on the branch node refer to bootstrap value of the phylogenetic tree; 500 bootstrap replicates were performed. Dots denote the four Arabidopsis COST proteins, with COST1 shown in magenta. Fragments per kilobase of transcript per million mapped reads (FPKM) values of the four Arabidopsis COST genes are shown on the right based on the RNA-seq in this study. Protein IDs are listed after abbreviated species names. The full species names are (from top to bottom): Arabidopsis thaliana, Arabidopsis lyrata, Capsella rubella, Brassica rapa, Populus trichocarpa, Solanum lycopersicum, Medicago truncatula, Aquilegia coerulea, Brachypodium distachyon, Oryza Sativa, Zea mays, Amborella trichopoda, Picea abies, and Physcomitrella patens. Triangles represent hidden subtrees, which are shown in detail in SI Appendix, Fig. S2. (C) Analysis of expression of the Arabidopsis COST genes by qPCR. (D) Pleiotropic defects of a cost1 mutant. Four-week-old and 45-d-old WT and cost1 plants are shown in the Upper and Lower, respectively.

Eight of the 12 angiosperm species examined in our study have only one copy of the COST1 gene. Due to the repeated polyploidization events in flowering plants, the majority of genes have ≥1 duplicate (39). In addition, genes with housekeeping functions that are subject to strong selective pressure tend to be maintained as single copies across species (40). Thus, the conserved phylogenetic distribution of the low copy number of COST1 suggests that it is subject to strong selection to retain one copy. In four species with COST1 duplicates, three (Brassica rapa, Populus trichocarpa, and Zea mays) have relatively recent whole-genome duplication events (41). A. thaliana has three closely related COST1-like genes (COST2 to -4) that were derived from duplication events after the divergence between A. thaliana and Arabidopsis lyrata lineages (Fig. 1B). COST2 and -4 are both truncated at the latter halves of the genes compared to COST1. COST3, on the other hand, shares similarity over its entire length of COST1 but has a premature stop codon. Together with our findings that the transcripts of COST2/4 and COST3 are undetectable or at an extremely low level, respectively, by RNA sequending (RNA-seq) and qPCR when compared with COST1 (Fig. 1 B and C), these findings suggest that COST2 to -4 may be pseudogenes.

To characterize the function of COST1 in Arabidopsis, a T-DNA insertion line, SALK_064001, was obtained from the Arabidopsis Biological Resource Center. Genomic PCR showed that the insertion in cost1 is homozygous, and qPCR indicated complete loss of transcript in the mutant background (SI Appendix, Fig. S3 A–C). The cost1 mutation dramatically affected plant growth; compared with WT (wild-type, Columbia-0) plants, cost1 has smaller leaves and reduced plant height (Fig. 1D and SI Appendix, Fig. S3 D and E). Microscopy analysis of the leaves indicated that the reduced plant size is caused by reduced cell size (SI Appendix, Fig. S3 F and G). The mutant also has dark green leaves that could be attributed to increased chlorophyll per equivalent fresh weight (SI Appendix, Fig. S3H).

cost1 Mutant Has Increased Drought Tolerance.

Based on the observation of the small, dark green phenotype of the cost1 mutant, we hypothesized that cost1 may be stressed and function in ABA or stress responses. We first tested the drought response of cost1 by withholding water for 2 wk and found that the cost1 mutant is more resistant to drought than WT plants (Fig. 2A and SI Appendix, Fig. S4A). To confirm that the cost1 phenotype is caused by disruption of the COST1 gene, the COST1 genomic sequence including 670 bp upstream of the start codon, the full-length COST1 coding sequence, and 540 bp downstream was introduced into the cost1 mutant background by Agrobacterium-mediated floral dip (42). The complementation lines generated expressed the COST1 gene, and were identical to WT in growth, drought tolerance, and water loss assays (SI Appendix, Fig. S4).

Fig. 2.

Knockout of COST1 confers drought tolerance and induces the expression of a spectrum of stress-responsive genes. (A and B) WT, cost1 (cost1-1), gCOST1#1 (complementation line with COST1 genomic DNA), and two COST1 RNAi lines (COST1-RNAi#1 and COST1-RNAi#3) were subjected to drought treatment for 2 wk, and the water loss of detached rosette leaves from each genotype was recorded every 30 min for 5 h (C). (D) Expression of representative drought responsive genes in WT and cost1 with and without drought treatment was assessed by qPCR. (E) Clustering of DEGs (fold-change in expression level >2 between two samples) in WT and cost1 plants with and without dehydration treatment, “C” denotes control and “D” denotes dehydration. Color legend denotes normalized gene-expression value. (F and G) Comparison of drought-regulated DEGs in whole-transcriptome RNA-seq. (H and I) Enrichment of Biological Process GOs in drought-regulated DEGs. Color scale: −log₁₀ (adjusted P value). Over- and underrepresentation are shown in red and blue, respectively.

To further demonstrate that the mutant phenotype is correlated with the COST1 expression level, we generated RNAi-COST1 lines with reduced COST1 expression (SI Appendix, Fig. S5A). RNAi-COST1 lines showed drought resistance and lower water loss than WT, similar to the phenotype observed in the cost1 mutant (Fig. 2 B and C). In addition, the RNAi lines and cost1 mutant had a smaller stomatal aperture and higher proline content than WT (SI Appendix, Fig. S5 B–D), which further indicates that COST1 is a negative regulator of drought responses. The expression patterns of nine dehydration-induced genes—RD29A, ABI2, ABI5, PP2C, RD22, COR15A, KIN1, COR414-TM1, and LTP3 (43, 44)—were tested by qPCR in WT and cost1 plants with or without dehydration treatment. As constitutively higher expression of these genes was observed in unstressed cost1, when compared with WT, with even higher expression in dehydrated cost1 plants, this supports a role for COST1 in drought responses (Fig. 2D).

Disruption of COST1 Causes Broad Induction of Stress-Responsive Genes.

To reveal the underlying mechanism of COST1 function in the drought response, we examined the transcriptome of 10-d-old WT and cost1 plants under control and dehydration conditions by RNA-seq (Fig. 2E). Statistically differentially expressed genes (DEGs; fold-change > 2) were evaluated and a gene ontology (GO) enrichment analysis supported the role of COST1 in drought tolerance, as many stress-associated pathways were significantly enriched in the gene set (Fig. 2 F–H). Among these drought up-regulated genes, 52 genes were constitutively induced in the cost1 mutant background when compared with WT (Fig. 2F). Importantly, 86 genes were constitutively up-regulated and 20 genes were constitutively down-regulated in the cost1 mutant, even without dehydration treatment (Fig. 2 F and G). The up-regulated gene set is highly enriched in water deprivation, ABA, salt, and cold pathway functions, while there is significant enrichment in photosynthesis-related pathways in the down-regulated genes (Fig. 2 H and I). This transcriptome analysis was consistent with the phenotype (increased drought tolerance and reduced plant growth) we observed in cost1 (Fig. 2 A and B), indicating an important role for COST1 in drought responses.

Drought-Mediated COST1 Protein Relocation and Its Degradation through the 26S-Proteasome and Autophagy.

Many stress-related genes are transcriptionally regulated (45). To test the effect of different stresses on COST1 transcript levels, 10-d-old WT seedlings were treated with NaCl, ABA, mannitol, or drought and the expression of COST1 was assessed by qPCR. Unexpectedly, no significant difference was found under any stress tested when compared with control conditions (SI Appendix, Fig. S6A). In addition, transgenic Arabidopsis lines were generated expressing GUS driven by the COST1 promoter. Ten-day-old transgenic seedlings harboring ProCOST1-GUS were stained for GUS activity, confirming that COST1 expression does not change significantly in response to drought (SI Appendix, Fig. S6B).

We then hypothesized that drought could potentially affect the stability of the COST1 protein, instead of its transcript level. To test this hypothesis, transgenic plants expressing COST1-YFP under a constitutive CaMV 35S promoter in the cost1 mutant background were generated. While under control conditions the YFP signal appeared throughout the cytoplasm, fluorescent puncta were observed when COST1-YFP transgenic plants were subjected to dehydration (Fig. 3 A and B), indicating a change in localization. There was also a significant reduction in the fluorescence intensity of COST1-YFP after dehydration when compared with control samples (Fig. 3 A and C). In addition, COST1-YFP protein degradation was observed over a time course of dehydration treatment; immunoblotting analysis detected an ∼30% decrease in COST1-YFP when treated for 6 h (Fig. 3D).

Fig. 3.

Expression of COST1 during drought stress. (A) Localization and characteristics of COST1-YFP protein after 3 h of dehydration treatment. (Scale bar, 10 µm.) (B and C) Quantification of fluorescence intensity and punctate structures before and after 3 h of dehydration treatment. Signals were quantified for at least 10 images per replicate, with 3 biological replicates. Asterisks indicate significant differences, compared with no treatment. (D) Immunoblot of COST1-YFP protein after dehydration treatment for the indicated times using antibodies against GFP. The number below indicates the band intensity of COST1-YFP, and Ponceau staining was used as a loading control. (E) Immunoblot of COST1-YFP protein after treatment with DMSO (control), MG132, or ConcA. (F) COST1-YFP was immunoprecipitated after treatment with DMSO, MG132, or ConcA under normal conditions, followed by detection using anti-GFP antibodies. (G) Ubiquitination of COST1-YFP after dehydration of 10-d-old COST1-YFP or YFP transgenic plants for 6 h. After immunoprecipitation with GFP-trap, samples were immunoblotted using antibodies against ubiquitin. Transgenic COST1-YFP plants were generated in cost1 mutant background with full complementation.

The primary general mechanisms for protein degradation in plant cells are via the 26S proteasome pathway (46) or by vacuole-mediated bulk protein degradation, generally through the autophagy pathway (25). To assess whether the 26S proteasome or autophagy are required for COST1 degradation, COST1-YFP seedlings (cost1-1 mutant background, with full complementation) were incubated with the proteasome inhibitor MG132 or the vacuolar degradation inhibitor ConcanamycinA (ConcA) over a time course of dehydration treatment. COST1-YFP protein accumulated after the dehydration treatment upon MG132 or ConcA incubation, compared with the DMSO (dimethylsulfoxide) control (Fig. 3 E and F). Immunoprecipitation using GFP-Trap demonstrated that COST1-YFP was ubiquitinated, with more COST1-YFP ubiquitination detected upon dehydration compared with the control (Fig. 3G and SI Appendix, Fig. S7). These results indicate that dehydration can induce the ubiquitination of COST1, which is then subjected to degradation by both the 26S proteasome and autophagy.

Colocalization of COST1 with ATG8 and NBR1.

To provide insight into the potential function of COST1, we assessed in more detail its subcellular localization. Upon transient expression of COST1-YFP in Arabidopsis protoplasts, punctate structures were observed. However, no colocalization was seen when we coexpressed COST1-YFP with organelle markers, including peroxisome (mCherry-peroxisome), Golgi (mCherry-Golgi), and PVC markers (mRFP-VSR2 and mCherry-Rha1) (47, 48) (SI Appendix, Fig. S8). Interestingly, constantly moving dots are observed in COST1-YFP transgenic plants upon dehydration treatment (Movie S1); we hypothesized that these dots may be autophagosomes, given the critical role of autophagy in the drought response (30, 31).

To test this hypothesis, we cotransformed protoplasts with COST1-YFP and mCherry-ATG8e, a marker that labels both autophagosomes in the cytoplasm and autophagic bodies after delivery to the vacuole. We found that the two proteins colocalized (Fig. 4 A, Upper), suggesting that COST1-YFP is found in autophagosomes or autophagic bodies. In plants, the selective autophagy adaptor NBR1 has been reported to interact and colocalize with ATG8 (28, 49). NBR1 specifically binds to ubiquitinated proteins and recruits them to autophagosomes for degradation (23, 27). COST1-YFP also colocalized with mCherry-NBR1 puncta (Fig. 4 A, Lower), which are likely to correspond both to autophagosomes and to ubiquitinated protein aggregates.

Fig. 4.

Direct interaction between COST1 and ATG8e. (A) Colocalization of COST1-YFP with mCherry-ATG8e and mCherry-NBR1. (Scale bar, 10 µm.) (B) Split luciferase analysis of the interaction between COST1-nLUC and cLUC-ATG8e. Different combinations of GFP-nLUC and cLUC-GFP with and without COST1 or ATG8e were used as negative controls. (C) GST pull-down assay between GST-COST1 and His-ATG8e. GST alone was used as negative control. (D) Co-IP of COST1-Flag with GFP-ATG8e. Agrobacterium-mediated coinfiltrations were carried out in tobacco leaves with combinations of 35S:GFP and 35S:COST1-Flag, and 35S:GFP-ATG8e and 35S:COST1-Flag. After 2 d of incubation, leaves were ground in liquid nitrogen and proteins immunoprecipitated with GFP-Trap. The immunoblot was probed with anti-GFP and anti-Flag antibodies. Leaves expressing 35S:GFP alone were used as a negative control.

Direct Interaction of COST1 with ATG8e.

We assessed whether COST1 and ATG8 can directly interact using a split luciferase assay in tobacco leaves (Fig. 4B) (50). cLUC-ATG8e interacted with COST1-nLUC, while the negative controls cLUC plus nLUC, cLUC-ATG8e plus GFP-nLUC, and cLUC-GFP plus COST1-nLUC did not interact. Next, we performed pull-down assays to confirm this interaction using GST or a GST-COST1 fusion as bait to precipitate His-ATG8e protein (SI Appendix, Fig. S9). As shown in Fig. 4C, GST-COST1 protein bound specifically to His-ATG8e but GST alone did not. Finally, coimmunoprecipitation (co-IP) analysis after transient expression in tobacco leaves also showed that COST1-Flag could specifically precipitate GFP-ATG8e but not the GFP protein alone (Fig. 4D).

Drought Tolerance of cost1 Is Dependent on Autophagy.

As COST1 is a previously uncharacterized gene involved in drought tolerance, we determined its genetic relationship to existing drought signaling pathways. ABA and H₂O₂-dependent signaling pathways play a major role in regulating drought responses. To test whether COST1 functions in ABA signaling, cost1 was crossed with abi1-1C, a gain-of-function mutant in the ABA signaling pathway (51, 52); with ost1, an SnRK2 protein kinase-null mutant in ABA and H₂O₂ signal transduction (53); and with aba3, an ABA biosynthesis-deficient mutant (54). To test the relationship of cost1 to H₂O₂ signaling, cost1 was crossed with ghr1 (18), another H₂O₂ signaling-deficient mutant. As shown in SI Appendix, Fig. S10 A–D, in a water loss assay, each of the double mutants had phenotypes intermediate between the corresponding single mutants; this suggests that COST1 acts independently of ABA and H₂O₂ signaling pathways. In addition, no significant difference from WT was observed in seed germination when cost1 mutant seeds were germinated on half MS (Murashige and Skoog) media supplemented with ABA (SI Appendix, Fig. S11), also suggesting that COST1 functions independently of ABA signaling.

Autophagy is required for drought tolerance in Arabidopsis, and autophagy-deficient mutants are hypersensitive to drought (30). In addition, COST1 colocalizes with autophagosome markers and directly interacts with ATG8e (Fig. 4), indicating a potential relationship to autophagy. To test the relationship between cost1 and the autophagy pathway, cost1 was crossed with atg5-1 and atg7-2 as representative autophagy-deficient mutants (55, 56). In contrast to ABA and H₂O₂ signaling pathway mutants, cost1 atg5-1 and cost1 atg7-2 double mutants have similar water loss and drought-sensitivity phenotypes to the atg5-1 or atg7-2 single mutants, suggesting that autophagy acts downstream of cost1 in drought tolerance (Fig. 5 A–C), and that a functional autophagy pathway is required for drought tolerance in the cost1 mutant. We also crossed cost1 with nbr1, a mutant defective in the NBR1-selective autophagy adaptor; nbr1 is unable to recover the cost1 low water loss phenotype (SI Appendix, Fig. S10E), suggesting that nbr1 is not involved in cost1-mediated drought regulation.

Fig. 5.

Autophagy is required for drought tolerance of the cost1 mutant. (A) WT, cost1, and the two cost1 atg double mutants after exposure to drought for 2 wk. (B and C) Water loss upon drought treatment of cost1 in combination with atg5-1 and atg7-2. Three independent experiments were done with similar results. Values are means ± SE of three replicates and at least 10 leaves from each genotype were assessed per replicate. Asterisks indicate significant difference. (D) Ten-day-old seedlings of the indicated genotypes were treated with or without 300 mM mannitol for 6 h, stained with MDC, and elongation zones of the roots were observed by epifluorescence. (Scale bar, 50 µm.) (E) Autophagosomes from D were quantified for at least 10 images per replicate, with 3 biological replicates. Asterisks indicate significant differences.

As autophagy is also required for tolerance of other stresses, we tested seedling sensitivity to salt. The cost1 mutant showed a higher survival rate than WT when 150 mM or 200 mM NaCl was included in the growth media (SI Appendix, Fig. S12), consistent with a function in autophagy. The role of COST1 in regulating salt and other stress responses warrants further exploration.

cost1 Mutants Have Increased Basal Autophagy.

To assess autophagosome formation in cost1 and cost1 atg double mutants, osmotic stress was introduced by treating plants with mannitol, a condition known to induce autophagy (30). Roots from plants of various genotypes were stained with MDC (monodansylcadaverine), a fluorescent dye that labels acidic vesicles, primarily autophagosomes, in vivo (57). Upon treatment with mannitol, an increased number of fluorescent puncta were seen in both WT and the cost1 mutant when compared with control conditions, and as expected this increase was absent in mutants in which ATG5 or ATG7 were disrupted (Fig. 5 D and E), suggesting that the fluorescent puncta correspond to autophagic structures. Even in the absence of stress, numerous fluorescent puncta were evident in the cost1 mutant and RNAi lines (SI Appendix, Fig. S13), suggesting constitutive activation of the autophagy pathway, and this was completely blocked in cost1 atg5-1 and cost1 atg7-2 double mutants (Fig. 5 D and E). This indicates that ATG5 and ATG7 are required for the potential constitutive autophagy seen in cost1, and again positions autophagy downstream of cost1.

To confirm the phenotype we observed by MDC staining, cost1 was crossed with a well-characterized autophagosome marker line, GFP-ATG8e (58). Confocal microscopy analysis suggested that the production of GFP-labeled autophagosomes was highly induced under nonstressed conditions in the cost1 mutant background compared with the WT control (Fig. 6A). Next, we used the release of free GFP to detect autophagic transport and degradation of GFP-ATG8e in the vacuole (58). Immunoblot analysis indicated more free GFP in cost1/GFP-ATG8e than in WT/GFP-ATG8e both before and after drought treatment (Fig. 6B and SI Appendix, Fig. S14), demonstrating that the increase in autophagosome number in the cost1 mutant is due to increased autophagosome formation and not decreased degradation. Autophagic flux is therefore higher in the cost1 mutant than in WT plants.

Fig. 6.

COST1 is a negative regulator of autophagy. (A) Confocal microscopy analysis of autophagy in the same autophagosome marker line GFP-ATG8e in the genetic background of WT, cost1, and COST1 overexpression lines #35 and #42. (Scale bar, 50 µm) Ten-day-old seedlings were treated with dehydration for 6 h or starvation for 16 h and representative images are shown. (B) Analysis of GFP-ATG8e cleavage as an indicator of autophagy activity in WT, cost1, and two COST1 overexpression lines with and without 6-h dehydration. Ponceau staining was used as a loading control. (C) Assay for drought tolerance of WT, cost1, and two COST1 overexpression lines. Four-week-old plants of the indicated genotypes were subjected to water withholding for 2 wk and a representative image is shown. (D) A working model of COST1 function in the drought response. Under normal growth conditions, COST1 inhibits stress responses by directly interacting with ATG8, leading to degradation and thus favoring plant growth. In stress conditions, COST1 proteins are degraded by both the 26S proteasome and autophagy, releasing ATG8, and thus the repression of autophagy, and in turn conferring drought tolerance.

Overexpression of COST1 Leads to ATG8e Degradation.

To assess the effect of increased COST1 protein on autophagy activity, we overexpressed COST1 in the same GFP-ATG8e marker line. RT-PCR (reverse-transcriptase PCR) indicated that the COST1 gene is highly expressed in all three independent transgenic lines tested (SI Appendix, Fig. S15A), and two independent overexpression lines, 35S:COST1 #35 and #42 were chosen for further study. Note that these lines express much higher levels of COST1 than the lines used for YFP-COST1 imaging and localization. First, confocal microscopy was employed to assess the effect of COST1 overexpression on autophagy. Compared with the GFP-ATG8e signal in the absence of COST1 overexpression, the fluorescence corresponding to GFP-ATG8e upon COST1 overexpression was almost undetectable (Fig. 6A). Next, by immunoblotting using GFP antibodies, we analyzed the cleavage of GFP-ATG8e to produce free GFP, as an indicator of flux through the autophagy pathway. A substantial reduction in both full-length GFP-ATG8e and free GFP was seen (Fig. 6B and SI Appendix, Fig. S16), and this reduction was more evident upon dehydration, consistent with our observations by confocal microscopy (Fig. 6A).

If GFP-ATG8e was simply degraded in the vacuole by autophagy upon COST1 overexpression, we would expect an increase in free GFP in the cleavage assay, which was not observed. To assess whether this effect was specific to the GFP-ATG8e transgene-encoded protein, we detected endogenous ATG8 with commercially available ATG8 antibodies, which detect most or all of the ATG8 isoforms in Arabidopsis. The overall ATG8 protein amount was found to be reduced in the COST1 overexpression lines (SI Appendix, Fig. S17). Meanwhile, compared with WT plants, COST1 overexpression did not significantly affect the ATG8e transcript level, as determined by RT-PCR (SI Appendix, Fig. S15A), suggesting that the decrease in ATG8e protein occurs posttranscriptionally, possibly due to degradation.

As ATG8 is a key factor in autophagosome formation, a reduction in the amount of ATG8 could potentially reduce the capacity of the autophagy pathway. We hypothesized that GFP-ATG8e protein might be degraded via the proteasome, thus regulating its availability to function in autophagy. To test this, the COST1 overexpression plants were treated with the proteasome inhibitor MG132 over a time course of 1 to 4 h. A substantial accumulation of GFP-ATG8e was seen after inhibitor treatment (SI Appendix, Fig. S15B), indicating that GFP-ATG8e is degraded by the proteasome upon COST1 overexpression. In line with a reduction in autophagy, the COST1 overexpression transgenic lines are more sensitive to drought and have a higher rate of water loss when compared with WT (Fig. 6C and SI Appendix, Fig. S15C).

During an extended length (>60 d) of short-day growth of WT and cost1, we observed a delay of leaf senescence and flowering in cost1 mutant (SI Appendix, Fig. S18). Autophagy is strongly associated with nutrient deficiency; to confirm the role of COST1 in autophagy, we thus assessed the effect of increased and decreased COST1 on autophagy induction upon carbon and nitrogen starvation. As shown in Fig. 6A, starvation strongly activated autophagy in WT and cost1, as evidenced by autophagosome accumulation; but autophagy is substantially reduced when COST1 is overexpressed. These results indicate that COST1 negatively regulates autophagy in both drought and starvation conditions. COST1 is therefore a key factor controlling the tradeoff between growth and stress responses that acts via the regulation of autophagy.

Discussion

Our phylogenetic analysis showed that COST proteins are plant-specific and broadly distributed in all plant species, including mosses. Disruption of the Arabidopsis COST1 gene caused profound defects in plant growth; the cost1 mutant is much smaller in size, implying that COST1 has a pivotal role in growth and cell expansion. The cost1 mutant also showed a strong drought resistant phenotype, with constitutively higher expression of stress-responsive genes, even under normal conditions, suggesting that COST1 regulates the balance between growth and stress resistance. COST1 is annotated as a myosin-4 like protein in the latest update from Araport; however, COST1 does not show sequence similarity to either of the typical plant myosins, which belong to classes VIII or XI, or to myosin-4 in animals. The basis for the annotation is therefore unclear.

Four closely related COST proteins were found in A. thaliana but a significant phenotype was observed by disrupting COST1 alone (Fig. 1 B and D). Meanwhile, only one copy of COST1 was found in A. lyrata and several other plant species analyzed (Fig. 1B). Considering the extremely low transcript level of COST2, COST3, and COST4 in our analysis (Fig. 1 B and C), the absolute read count of close to zero for these three genes in the TraVA database (http://www.travadb.org/), and the prediction that these genes would not encode full-length proteins, it is likely that these three genes are pseudogenes.

Over the last few decades, key components in plant growth and stress responses have increasingly been found to be regulated by the ubiquitin-proteasome pathway (UPS) (5, 59, 60). Compared with animals, many more protein quality control-related genes are found in the genomes of plant species, including unfolded protein response and proteasome pathway genes (61). For example, ∼2,000 E3 ligase and UPS genes are predicted to exist in the genome of Arabidopsis, outnumbering those in animals (46). In this study, ubiquitination and the UPS were shown to regulate the abundance of the COST1 protein (Fig. 3). This suggests that proteins may exist that act as E3 ligases in the degradation of COST1, although the identity of these factors remains to be discovered.

In addition to the UPS, autophagy can also efficiently dispose of unwanted proteins, protein aggregates, or organelles (25, 26). In plants, autophagy functions in the response to many stresses, including drought, salt, and oxidative stresses (30, 31). Whereas the ABA and H₂O₂-dependent signaling pathways have been widely studied as responsible for drought tolerance, genetic evidence showed that the increased drought resistance of cost1 is independent of both of these pathways (SI Appendix, Fig. S10 A–D). Instead, our genetic analysis clearly indicates that atg5-1 and atg7-2 are epistatic to cost1 in drought tolerance (Fig. 5), demonstrating that autophagy is required for the cost1 mutant drought resistance. The cost1 mutant has constitutively activated autophagy, and COST1-YFP interacts with ATG8e and is recruited to autophagosomes upon drought stress. These data suggest that COST1 both regulates autophagy and is also degraded by the autophagy pathway, a feature shared with the core autophagy regulator ATG1 (62). However, whereas ATG1 is a positive regulator of autophagy, and its degradation results in a balancing feedback loop to allow moderation of the extent of autophagy activation, COST1 is a negative regulator, and its degradation results in positive feedback, allowing rapid response to stress conditions and therefore promoting plant survival.

Many ATG8-interacting proteins act as selective autophagy adaptors and share a conserved LIR (LC3-interacting region) motif (63). A search for LIR motifs using the online iLIR tool (http://repeat.biol.ucy.ac.cy/iLIR/) (64) finds no clear LIR motif in COST1, suggesting that COST1 interacts with ATG8 via a different mechanism. Interestingly, the 26S proteasome core subunit RPN10, which acts as a receptor for selective autophagy of the proteasome, has a distinct motif, termed a UIM (ubiquitin-interacting motif), for recognition by the UIM docking site of ATG8 (65). Narrowing down the specific region of COST1 that interacts with ATG8 will be an interesting next step in dissecting the underlying mechanism of interaction.

We propose a working model for the function of COST1 in drought responses (Fig. 6D). Through direct interaction with ATG8, COST1 inhibits autophagy under normal conditions, thus favoring plant growth. Upon drought stress, COST1 is ubiquitinated and subjected to degradation by the UPS and autophagy. The subsequent decrease in COST1 protein levels releases ATG8, allowing autophagosome formation and promoting drought tolerance, while inhibiting growth. TOR (target-of-rapamycin) is a negative regulator of autophagy in plants (66), as in animals and yeast, but many of the upstream regulators of TOR are absent from plants. Plant-specific proteins like COST1 may provide alternative mechanisms of regulating autophagy and growth in plants. The broad distribution of COST genes in land plant species suggest that COST family proteins are functionally conserved, raising the possibility of manipulating COST-like genes to confer drought tolerance in agricultural species (67).

Materials and Methods

Plant Materials and Growth Conditions.

All A. thaliana lines used in this study are in the Columbia-0 (Col-0) accession. Five-day-old seedlings grown on half strength MS medium (Sigma-Aldrich) containing 1% (wt/vol) sucrose and 0.8% (wt/vol) agar were transplanted into soil, then kept under 16 h of light and 8 h of dark in a growth chamber at light/22 °C or dark/18 °C. The cost1-1 T-DNA insertion (cost1, SALK_064001) was confirmed by genomic PCR using the T-DNA left-border primer LBa1 and two COST1 gene-specific primers (Dataset S1). Disruption of COST1 gene expression in a homozygous cost1-1 mutant was confirmed by RT-PCR and qRT-PCR (quantitative RT-PCR). ACTIN2 was used as an internal control.

Double mutants were obtained by crossing cost1-1 with nbr1 (Salk_135513) (28) and various drought-sensitive mutants, including abi1-1C (52), aba3 (SALK_054454), ost1(SALK_008068), ghr1 (SALK_031493), atg5-1 (SAIL_129_B07) (55), and atg7-2 (GK-655B06) (56). Genotyping was performed by sequencing for abi1-1C and by genomic PCR for the other mutants.

To introduce GFP-ATG8e into the cost1 mutant background, cost1 was crossed with a 35S:GFP-ATG8e transgenic line (68). Homozygous plants were confirmed by genomic PCR for cost1 and kanamycin resistance for GFP-ATG8e.

Vector Construction and Plant Transformation.

To generate the construct for cost1 mutant complementation, the genomic DNA sequence including 670 bp upstream of the predicted ATG start codon of COST1 and 540 bp downstream of the stop codon was amplified by PCR primers gCOST1F and gCOST1R (Dataset S1), and ligated into the pCAMBIA1300 binary vector after digestion with EcoRI and SalI.

To generate the constructs for COST1 overexpression, the ORFs were amplified using gene-specific primers (Dataset S1), and inserted into a modified pCAMBIA1300S vector (69) via BamHI and SpelI restriction sites. Two COST1 overexpression lines used are termed 35S-COST1-OX-#35 and -#42.

To construct the vector to knock down the expression of COST1 by RNAi, a pair of gene-specific primers, RNAi-COST1F and RNAi-COST1R, were used to amplify a cDNA fragment specific to COST1 (Dataset S1). The resultant fragment was further introduced into the modified binary vector pFGC5941 (ABRC Stock CD3-447) by a two-step cloning strategy, according to the manufacturer’s instructions (http://www.chromdb.org/rnai/vector_info.html). The two COST1 RNAi lines characterized in detail are termed COST1-RNAi-#1 and -#3.

To generate the COST1 promoter-GUS construct, 700 bp upstream of the COST1 coding region was amplified with COST1 promoter specific primers ProCOST1-F and PrCOST1-R (Dataset S1). The 700-bp PCR fragment was digested with BamHI and SmaI and inserted into the pCAMBIA1381Z vector (70). Histochemical analysis by GUS staining was described previously (69).

All constructs were introduced into Agrobacterium strain GV3101 and then into Arabidopsis plants by the floral-dip method (42). Transgenic plants were screened on half-MS medium supplemented with 50 mg/L hygromycin and resistant transformants were selected. Transgene expression was assessed in homozygous T₃ progeny.

Drought Treatment and Water Loss Assay.

To assess drought tolerance, 5-d-old seedlings from different genotypes transplanted to soil were grown for 3 wk. Plants were then subjected to progressive water stress by withholding water for another 2 wk, and images were taken. Equal numbers of plants of different genotypes were grown in the same tray to minimize experimental variation. Measurement of stomatal aperture was based on the ratio of width/length and described in detail previously (71).

To assess water loss, 10 leaves per individual plant of 4-wk-old WT, mutant, complementation, and RNAi genotypes were excised, and fresh weights were determined at the designated time intervals. Three replicates were performed. Water loss was represented as the percentage of the initial fresh weight at each time point.

GST Pull-Down Assay.

For GST-COST1 protein purification, the COST1 cDNA was inserted into pGEX-4T1 via the EcoR I and SalI sites. After sequence confirmation, the construct was introduced into Escherichia coli strain BL21 (DE3). GST-COST1 protein synthesis was induced with 0.3 mM IPTG at 16 °C for 18 h. Purification of GST-COST1 was performed with Glutathione Sepharose 4B beads (GE Healthcare) following the manufacturer’s protocol.

To test the interaction between COST1 and ATG8e, the full-length coding sequence of ATG8e was subcloned into the pET-28a (+) vector and expressed in BL21 cells to produce a His-ATG8e fusion protein. A half-microgram of GST-COST1 or GST alone as control was incubated with glutathione resin (GE) at 4 °C for 1 h with shaking, then 0.5 µg His-ATG8e was added. The incubation was continued overnight and the beads were washed several times. The beads were boiled in 2× SDS loading buffer and proteins analyzed by immunoblotting using anti-His (M20001; Abmart) and anti-GST (M20007; Abmart) antibodies.

BiFC Assay.

BiFC assays were performed in Nicotiana Benthamiana leaves, as described previously (69). The firefly LUC enzyme was divided into N-terminal (nLUC) and C-terminal segments (cLUC), and COST1 and ATG8e were separately fused with nLUC and cLUC. The resulting constructs, cLUC-ATG8e, COST1-nLUC, and vector controls were each introduced into Agrobacterium strain GV3101. After growth in LB (Luria-Bertani) liquid medium, cells were collected and resuspended in infiltration buffer (10 mM MES, 10 mM MgCl₂, and 0.15 mM acetosyringone, pH 5.8) to OD₆₀₀ = 1.0 (50). Equal volumes of various combinations of Agrobacterium strains were mixed and coinfiltrated into N. benthamiana abaxial leaves with a needleless syringe. After the infiltration, plants were incubated at 22 °C for 3 d and infiltrated with 1 mM luciferin before observation (50, 72).

Coimmunoprecipitation.

Agrobacterium tumefaciens harboring 35S:GFP and 35S:COST1-Flag or 35S:GFP-ATG8e and 35S:COST1-Flag constructs were coinfiltrated into N. benthamiana leaves. Three days after the inoculation, leaves were harvested and quickly frozen in liquid nitrogen. Co-IP buffer (5 mM ethylenediamine tetracetic acid, 100 mM NaCl, 50 mM Tris⋅HCl, 10% glycerol, 0.2% Nonidet P-40, 0.1% Triton X-100, 50 µM MG132, and complete protease inhibitor mixture tablet, pH 7.5) was used to extract the proteins and the supernatant was centrifuged two times at 18,000 × g for 15 min (50). GFP-Trap beads (10 µL; ChromoTek) were added to 50 mL protein extract and incubated overnight. GFP-Trap beads were then washed three times with co-IP buffer. Bound proteins were released by adding 2× protein loading buffer and boiled for 5 min. For immunoblotting analysis, monoclonal anti-GFP (Sigma-Aldrich) and monoclonal anti-FLAG M2 (Sigma-Aldrich) antibodies were used to detect the relevant fusion proteins (72).

Confocal Microscopy.

To determine the subcellular localization of COST1, 35S:COST1-YFP was transiently expressed in Arabidopsis protoplasts (73) and analyzed using a confocal microscope at 520-nm and 550-nm wavelengths for excitation and emission (Zeiss LSM 510 META). The empty pA7-YFP vector served as control. Ten-day-old transgenic COST1-YFP plants (in cost1-1 mutant background) were used to characterize the COST1-YFP protein localization in response to dehydration. Relative fluorescence intensity was quantified by using imageJ software, the ratios of fluorescence intensity are shown, and signals from YFP alone transgenic plants were set as 1.0.

For colocalization analysis, mCherry was fused to the N terminus of ATG8e or NBR1 in the pAN583 vector. Combinations of equal concentrations of plasmids encoding COST1-YFP with mCherry-ATG8e, mCherry-NBR1, or other organelle markers were used for cotransformation. Transiently transformed Arabidopsis protoplasts were analyzed by confocal microscopy at excitation and emission wavelengths of 520 and 550 nm for YFP, 584 and 607 nm for mRFP, and 575 and 650 nm for mCherry.

MDC Staining.

MDC staining of mannitol-treated Arabidopsis roots was described previously (74). Briefly, 7-d-old seedlings from different genotypes were treated with half MS liquid medium (as the control) or 300 mM mannitol in half MS liquid medium for 6 h, followed by incubation with 0.05 mM MDC for 10 min in the dark. After three brief washes with 1× PBS, samples were observed by epifluorescence microscopy (Carl Zeiss), using a DAPI-specific filter to visualize MDC fluorescence.

Seed Germination Assay.

The seed germination assay was carried out based on previous studies (69, 70, 75). To minimize experimental variability, cost1 and WT plants were grown side by side and harvested at the same time. Seeds were sown on the same plate containing MS medium with 2% sucrose and 0.8% agar, supplemented without or with different concentrations of ABA as indicated. Plates were stratified at 4 °C for 3 d and moved to 22 °C with 16-h light and 8-h dark cycles in a growth chamber. Germination greening is defined as the cotyledons that have clearly expanded and turned green. About 100 seeds of cost1 and WT were used for each experimental treatment, and three biological replicates were performed for statistical analyses.

Identification of COST1 Homologs and Their Phylogenetic Analysis.

The amino acid sequences of COST1 homologs in A. thaliana and other plant species were obtained by using the COST1 protein sequence as query in BLASTp (76) searches against TAIR10 (https://www.arabidopsis.org/) and Phytozome database v12 (https://phytozome.jgi.doe.gov/pz/portal.html) annotated protein sequences, respectively. A match with an E-value ≤ 1e-5 was treated as a potential COST1 homolog. These potential homologs were then scanned for the presence of the DUF641 domains using the PF04859 hidden Markov model (HMM) from Pfam (77). Sequences with matches above the default threshold were regarded as COST1 homologs. From the COST1 homologs, the amino acid sequences of the DUF641 domains were aligned to the HMM using hmmalign as implemented in HMMER3 (78). The domain sequence alignments were used to infer the COST1 homolog phylogeny using PhyML v3.0 (79), with the following parameters: -b 500 -m JTT -f e -v e -c 4 -a e. MEGA7.0 (80) was employed for further visualization and editing of the phylogenetic tree. The full-length sequence alignment of protein homologs was created using the Clustalx1.83 program with default settings.

RNA Extraction and Real-Time PCR.

To analyze transcript levels, qRT-PCR was performed with RNA samples isolated from 10-d-old seedlings grown on half MS media harvested at the indicated times after exposure to 100 µM ABA, 200 mM NaCl, 400 mM mannitol, or drought conditions. Total RNA was extracted with TRIzol reagent (Takara) and treated with RNase-free DNase (Promega). The first strand cDNA was synthesized with the Go Script Reverse Kit (Promega). PCR was performed in 96-well optical reaction plates (CFX Connect) after preincubation for 5 min at 95 °C, followed by 40 cycles of denaturation at 95 °C for 15 s, annealing at 55 °C for 20 s, and extension at 72 °C for 30 s. Amplification of the detected genes was monitored every cycle by SYBR Green fluorescence. All experiments were performed in the CFX Connect real-time PCR detection system using an AceQ qPCR SYBR Green Master Mix (Vazyme Biotech). Results were normalized to the reference gene ACTIN2 using the ΔΔCt method (81). Each experiment was repeated three times. Primers used in this study are listed in Dataset S1.

RNA-Seq and Data Analysis.

WT and cost1 seeds were germinated and seedlings grown vertically on 100-mm × 100-mm square plates (Fisher Scientific) on half-strength MS medium for 10 d. Seedlings were dehydrated for 6 h. Samples were immediately ground and total RNA was extracted using an RNeasy Plant Mini Kit (Qiagen, Cat. No. 74904). RNA-seq libraries were prepared and subjected to paired end sequencing with read length 250 bp. Read counts in each library were normalized using the TMM method, which is the weighted trimmed mean of M-values proposed by Robinson and Oshlack (82), where the weights are from the Δ method on binomial data. Sequences were aligned to the Arabidopsis TAIR10 genome, using STAR (v2.4.0). The reads count for each annotated gene were calculated by htseq-count (v0.6.0 with parameters “-t mRNA -m intersection-nonempty–stranded no”; mRNA denotes messenger RNA). Differential gene expression between samples was assessed using negative-binomial generalized-linear models with DESeq2 (83, 84). A gene was considered as expressed in a sample if the average reads per kilobase and million mapped reads (RPKM) among replicates of the gene is >1, and a gene was considered as differentially expressed between samples if the fold-change between two average RPKM >2. GO functional category information was obtained from TAIR. GO enrichment analysis was conducted using Fisher’s exact test, and the P values were adjusted to account for multiple testing (85), and the corresponding processed data are listed in Datasets S2–S4.

Statistical Analysis.

A two-tailed Student t test was used for comparison of treatments with the control, and the Fisher’s least significant difference test was used for multiple mean comparisons.

Data Availability.

Raw sequencing data and the processed data for RNA-seq analysis were deposited in the National Center for Biotechnology Gene Expression Omnibus (GEO) under the accession no. GSE113515. Accession numbers for each gene are listed in Dataset S1. Sequence and detailed information for each gene can be found at TAIR (https://www.arabidopsis.org/). All other data are available within the paper and SI Appendix files.