Ceramides mediate positional signals in Arabidopsis thaliana protoderm differentiation

Sessile lifestyle imposes a bewildering array of biotic and abiotic stresses on vascular land plants. To defend against a hostile environment on land, plants have a specialized cell layer called the protoderm/epidermis at the interface between the plant body and the external environment. The protoderm/epidermis is a sheet-like tissue consisting of polarized protodermal/epidermal cells that synthesize and secrete surface lipids such as very-long-chain fatty acid (VLCFA) derivatives to form a protective hydrophobic layer known as the cuticle. The protodermal/epidermal cells adhere strongly to each other and protect the plant body from harmful invaders. Intact protodermal/epidermal cells are also crucial for shoot development. For example, loss of function of some Arabidopsis regulators involved in protodermal cell differentiation leads not only to a defective epidermis but also to a disorganized shoot apical meristem (SAM) structure (Abe et al., 2003; Johnson et al., 2005; Ogawa et al., 2015), indicating that intact protodermal cells communicate with inner cells to achieve the precise organization and function of the SAM.

Lateral root primordia (LRPs), which are post-embryonically generated from a subset of cells in the inner layer, develop the cuticularized outermost layer like other organs (Malamy and Benfey, 1997; Berhin et al., 2019; Torres-Martínez et al., 2019). De novo formation of the multi-layered structure of LRPs is carried out through tightly coordinated periclinal cell divisions. In the first periclinal cell division leading to Stage II LRPs, the plerome is laid down in the innermost layer, and in the second periclinal division to Stage III LRPs, the periblem is laid in the middle layer (Torres-Martínez et al., 2019). Subsequently, LRPs develop within this layered structure, and the outermost layer of Stage III LRPs eventually gives rise to the outermost layer of lateral roots (LRs), consisting of protodermal/epidermal cells and root cap cells. Thus, LRP development emerges as a good system with which to understand the molecular mechanism underlying the differentiation of protodermal/epidermal cells.

The protoderm/epidermis of aerial parts has previously been shown to be precisely marked by expression of ARABIDOPSIS THALIANA MERISTEM LAYER1 (ATML1) and its closest paralog PROTODERMAL FACTOR2 (PDF2), which encode HD-ZIP class IV transcription factors (TFs) (Lu et al., 1996; Sessions et al., 1999; Abe et al., 2003). ATML1 and PDF2 specifically bind to the L1 box, a well-conserved cis-regulatory element within the promoter region of shoot protoderm/epidermis-specific genes (Abe et al., 2001; 2003), and orchestrate a variety of protoderm/epidermis-specific events during shoot development. Thus, ATML1 and PDF2 act as master regulators of protodermal/epidermal cell differentiation in aerial parts (Abe et al., 2001, 2003; Ogawa et al., 2015), and dictate the location where the protodermal/epidermal identity manifests. Importantly, an L1 box is also found in the promoter of ATML1 and PDF2 (Abe et al., 2003). Consistent with this, the constitutive overexpression of ATML1 increases the transcript levels of endogenous ATML1 and PDF2 (Takada et al., 2013), and hypomorphic atml1-1; pdf2-1 mutants show greatly decreased transcript levels of ATML1 and PDF2 relative to single mutants or wild-type plants (Ogawa et al., 2015). A transcriptional feedback loop therefore directs the transcription of ATML1 and PDF2, although it remains puzzling how such autoregulation ultimately modulates the position-dependent expression of ATML1 and PDF2.

Here, we report that ATML1 and PDF2 function redundantly as crucial regulators for protodermal cell differentiation during LRP development. Our data show that the outermost layer-specific activity of ATML1 is greatly influenced by VLCFAs, which have recently been shown to be essential in various plant development processes (Bach et al., 2008; Markham et al., 2011; Trinh et al., 2019). Furthermore, we identify VLCFA-containing ceramides (VLCFA-Cers) as lipidic ligand candidates of ATML1 and demonstrate that VLCFA-Cers are essential components of the autoregulation loop controlled by ATML1 and PDF2. Based on our findings, we propose a novel molecular framework for the spatial control of protoderm/epidermis differentiation by a lipid-TF complex.

ATML1 and PDF2 play crucial roles in embryonic and shoot development (Abe et al., 2003; Ogawa et al., 2015). However, whether ATML1 and PDF2 are also involved in root development has not been clarified. Thus, we first re-evaluated root phenotypes in the hypomorphic atml1-1; pdf2-1 mutant.

In contrast to no apparent phenotype with the primary root including the radicles, the atml1-1; pdf2-1 double mutant showed severe defects in LR formation (Fig. 1A; Fig. S1). Because no significant defects in the density of LRs and LRPs in 10-day-old atml1-1; pdf2-1 seedlings were observed (Fig. S1), we next focused on LRP development. We determined the position of each LRP tip relative to the overlaying layers, which reflects the degree of progress of LRP development (Fig. 1B) (Malamy and Benfey, 1997; Casimiro et al., 2003), and evaluated the distribution of LRP tip locations in 10-day-old seedlings. The proportion of LRPs whose tips stayed at the pericycle layer was much higher in atml1-1; pdf2-1 than in each single mutant and wild type (Fig. 1C). Because obvious phenotypes were not observed in LRP development of each single mutant (Fig. 1A,C; Fig. S1), we concluded that ATML1 and PDF2 play redundant and essential roles in the early stages of LRP development as well as shoot development.

To further understand the molecular function of these HD-ZIP class IV TFs in LRP development, we generated transgenic Arabidopsis plants, in the atml1-3; pdf2-1 background, that expressed the enhanced green fluorescent protein (EGFP) fused to ATML1 (ATML1-EGFP) under the control of native ATML1 regulatory sequence (gATML1-EGFP; atml1-3; pdf2-1: Fig. S2) or the PDF2-EGFP under the control of native PDF2 regulatory sequence (gPDF2-EGFP; atml1-3; pdf2-1: Fig. S2), and examined the expression pattern of ATML1 and PDF2 during LRP development. Whereas no EGFP fluorescence was detected in LRPs whose tips were positioned in the pericycle layer (Fig. 2A), ATML1-EGFP and PDF2-EGFP signals were detected in the nuclei of the outermost cells of LRPs whose tips were positioned in the endodermis layer (Fig. 2B). Subsequently, the fluorescence of ATML1-EGFP and PDF2-EGFP was observed in the outermost cells of LRPs and young LRs (Fig. 2C-E). According to the conventional model, when LRPs reach the endodermis layer (Stage III/Stage IV; Malamy and Benfey, 1997; Casimiro et al., 2003), the cell fate of the protoderm is determined (Torres-Martínez et al., 2019). Thus, the spatial and temporal expression patterns of ATML1-EGFP and PDF2-EGFP strongly suggest that these two TFs are appropriate markers for protodermal and root cap cell differentiation, and play a crucial role in initiating protoderm and root cap differentiation during LRP development.

The above expression analysis revealed that ATML1 and PDF2 are expressed in the outermost cells during LRP development. However, our observations also showed that ATML1 and PDF2 are not expressed in the outermost cells after the first sloughing of root cap cells from LRs, even though these cells occupy the outermost cell position (Fig. 2F). This paradoxical observation led us to the idea that an outermost-cell position alone is not sufficient for the expression of ATML1 and PDF2, and that their expression may also depend on cell lineage. Indeed, when we analyzed the expression of ATML1 at other developmental stages and in other organs, stable ATML1 expression was observed specifically in clonal outermost cells derived from the cells expressing ATML1 at the one-cell embryonic stage (Takada and Jurgens, 2007), but was not detected in the outermost cells of the primary root after the loss of cell lineage (i.e. after sloughing of root cap cells from LRs) (Fig. S3).

In a lineage-based cell-fate decision mechanism, cell fate is restricted early in development and passed on to descendent cells (Scheres, 2001). Considering the autoregulation-dependent transcription of ATML1 and PDF2, ATML1 and PDF2 proteins can be passed on via cell division as cell-fate determinants, conferring the cell-lineage dependence of their expression. In a position-dependent cell-fate decision mechanism, by contrast, cell fate can be flexibly changed according to cell position (Scheres, 2001). Because the epidermal differentiation of inner cells has been shown to be strictly inhibited even when these cells are descendants of the epidermal cell (Stewart and Dermen, 1975), it seems likely that an unidentified candidate signal may modulate the inheritance or function of ATML1 and PDF2 proteins according to cell position.

To test this assumption, we used a heat-inducible transient expression system, by the HEAT SHOCK PROTEIN 18.2 (HSP) promoter (Takahashi et al., 1992), for the induction of ATML1^WT-EGFP and NLS-mCherry (HSP::NLS-mCherry; HSP::ATML1^WT-EGFP) in LR cells (Fig. 3A,B). Heat-pulse induction of the transcription of both ATML1^WT-EGFP and NLS-mCherry was confirmed by RT-PCR analysis (Fig. S4A). The ubiquitous expression of NLS-mCherry as a control indicated that transcription from the HSP promoter and subsequent translation was successfully induced in all LR cells (Fig. 3C,D; Figs S4B and S5A,B). ATML1-EGFP was also induced in all cells of LR immediately after induction (Fig. S4B). However, in contrast to mCherry, EGFP fluorescence was detected specifically in the nuclei of the outermost cells 2-3 h after heat induction, but not in those of the inner cells (Fig. 3C,D; Fig. S5A,B). Moreover, ATML1^WT-EGFP fluorescence was not observed in the outermost cells of LR after the first sloughing of root cap cells from LR (Figs S5A,E and S6). These results indicate the existence of an unidentified candidate signal that relays the information of cell position and lineage, affects ATML1 function, and is therefore required to maintain ATML1 expression.

HD-ZIP class IV family proteins, which are expressed predominantly in protodermal/epidermal cells, are characterized by a steroidogenic acute regulatory protein-related lipid-transfer (START) domain (Fig. 3A) (Ariel et al., 2007). Intriguingly, START domains from ATML1 and PDF2 have been shown to post-translationally stimulate TF activity via lipid-protein interactions in yeast (Schrick et al., 2014). To identify amino acids conserved among HD-ZIP class IV TFs, we compared the sequences of START domains of HD-ZIP class IV TFs with those of HD-ZIP class III TFs, which are also characterized by a START domain (Ariel et al., 2007), and identified several amino acids that are conserved only in HD-ZIP class IV TFs (Fig. S7A). In addition, to validate the importance of the identified amino acids in lipid binding, we compared the amino acid sequences of the START domains of ATML1 and PDF2 with the known ligand contact site of mammalian START domains (Fig. S7B) (Roderick et al., 2002; Schrick et al., 2014). Based on these sequence alignments, the W471 residue of ATML1 protein exhibits strong similarity with the ligand contact site of mammalian START domains (Fig. S7B). Thus, we focused on W471 as a putative lipid contact site in ATML1.

To examine the functional significance of the W471 residue in post-translational regulation of ATML1, we introduced a W471L point mutation into ATML1 based on a recent study (Schrick et al., 2014), and generated transgenic plants harboring the heat-inducible system of ATML1^W471L-EGFP and NLS-mCherry (HSP::NLS-mCherry; HSP::ATML1^W471L-EGFP) (Fig. 3A,B). In these plants, a single pulse of heat treatment induced the expression of ATML1^W471L-EGFP and NLS-mCherry in young LR immediately after heat-pulse treatment (Fig. S4C,D). In contrast to ATML1^WT-EGFP, transient induction of ATML1^W471L-EGFP in young LRs resulted in no fluorescence even in the outermost cells 2-3 h after heat induction (Fig. 3E,F; Fig. S5A,C). These observations indicate that W471 is crucial for the unidentified signal-mediated modulation of ATML1 protein in the outermost cells.

In order to screen for potential lipid species that interact with the ATML1 START domain as a component of the unidentified signaling pathway, we performed in vitro protein-lipid binding assays and found that the recombinant Glutathione S-transferase (GST) tagged START domain of ATML1 binds to VLCFA-Cers, ceramide-1-phosphates, phosphatidic acids, phosphatidylserines and several phosphatidylinositol phosphates (Fig. 4A; Fig. S8A,C). Interestingly, further evaluation of the effect of the W471L mutation revealed that the W471L mutation specifically attenuates the interaction between GST-START and VLCFA-Cers (Fig. 4A-C; Fig. S8A,B).

To determine whether the post-translational regulation of ATML1 is affected by VLCFA-Cers, we treated HSP::NLS-mCherry; HSP::ATML1^WT-EGFP seedlings with cafenstrole, an inhibitor of VLCFA biosynthesis (Nobusawa and Umeda, 2012). Consistently, 0.3 µM cafenstrole treatment strongly affected levels of VLCFA-containing sphingolipids, including ceramides (Fig. S9A-D). In HSP::NLS-mCherry; HSP::ATML1^WT-EGFP seedlings treated with 0.3 µM cafenstrole, the fluorescence of ATML1^WT-EGFP was significantly attenuated in the outermost cells of young LRs 2-3 h after heat induction (Fig. 5A,B; Fig. S5A,D), confirming that the stability of ATML1 protein is affected by VLCFA-Cers. Moreover, the same treatment strongly reduced both the fluorescence intensity of EGFP and the number of outermost cells expressing ATML1-EGFP in gATML1-EGFP; atml1-3; pdf2-1 seedlings (Fig. S10).

To further examine whether the modulation of ATML1 stability by VLCFA-Cers is involved in the autoregulated transcription of ATML1 and PDF2, we analyzed the link between sphingolipid metabolism and their mRNA levels. Consistent with the function of ATML1 and PDF2 as positive regulators of their own expression, cafenstrole treatment strongly reduced the amounts of ATML1 and PDF2 mRNAs (Fig. 5C). Fumonisin B1 (FB1), although it is known to induce programmed cell death, can inhibit the activity of ceramide synthases that preferentially use VLCFA as a substrate (Markham et al., 2011). Seedlings grown in the presence of FB1 at 0.5 µM, which is a moderate but effective concentration (Markham et al., 2011) (Fig. S9E-H), also greatly reduced the mRNA levels of both genes (Fig. S11).

To strengthen the idea that VLCFA-Cers are essential components of the autoregulated pathway that controls protodermal/epidermal cell differentiation in cooperation with ATML1, we analyzed the expression pattern of PASTICCINO2 (PAS2), an essential enzyme for VLCFA biosynthesis (Bach et al., 2008). First, we confirmed that transgenic Arabidopsis, which express PAS2-EGFP under the control of native regulatory sequence of PAS2 (gPAS2-EGFP), can rescue pas2-3 from embryonic lethality (Fig. S2), indicating that PAS2-EGFP expression reflects the pattern of endogenous PAS2 expression. The EGFP fluorescence was detected in the outermost cells of LRPs whose tips were positioned in the pericycle layer (Fig. 6A), suggesting that VLCFA production precedes the expression of ATML1 and PDF2. However, the expression pattern of PAS2-EGFP was very similar to that of ATML1 and PDF2 in LRs and other tissues (Fig. 6B-E; Fig. S12), consistent with the presence of an L1-box sequence in the PAS2 promoter (Nobusawa et al., 2013). In addition, a previous study has shown that gene silencing of ATML1 and PDF2 downregulates the mRNA levels of VLCFA biosynthesis-related genes, including PAS2 (Rombolá-Caldentey et al., 2014). These results suggest that VLCFA-containing sphingolipid biosynthesis is initiated before the expression of ATML1 and PDF2, but maintained in the outermost cells mainly via transcriptional activation of biosynthesis genes by a positive-feedback loop involving ATML1 and PDF2. Consistent with this, several sphingolipid pools in viable hypomorphic atml1-1; pdf2-1 plants showed a significant decrease in VLCFAs (Fig. 6F-I).

Sphingolipids are essential components of the cell membrane and are known to be embedded in lipid microdomains. Lipid microdomains play an important role in various cellular processes, including signal transduction and membrane trafficking (Simons and Ikonen, 1997; Simons and Toomre, 2000; Surma et al., 2012). In mammalian polarized epithelial cells, lipid microdomain-based apical sorting and trafficking results in sphingolipid-rich apical membrane compartmentation (Fig. 7A) (Simons and Ikonen, 1997; Simons and Toomre, 2000; Surma et al., 2012). Similarly, the plasma membrane of plant protodermal/epidermal cells in embryos, shoots and roots is also known to be highly polarized and compartmentalized (Fig. 7A-C) (Watanabe et al., 2004; Luo et al., 2007; Liao and Weijers, 2018; Nakamura and Grebe, 2018). Previous studies have demonstrated that some membrane proteins, such as the ATP-binding-cassette transporter WBC11, are located exclusively in the outer lateral membrane (Fig. 7B) (Luo et al., 2007), whereas we found that Remorin1.2 (REM1.2) (Jarsch et al., 2014) is localized to the inner lateral membrane of the LRP protodermal cells (Fig. 7C). To analyze the role of VLCFA-containing sphingolipids in the outer-inner polarity of plant LRP protodermal cells, we analyzed the localization of GFP-WBC11 and YFP-REM1.2 in LRP protodermal cells in presence of FB1. Surprisingly, the localization of GFP-WBC11 was affected in LRP protodermal cells of seedlings treated with FB1, but that of YFP-REM1.2 was not (Fig. 7B,C). Therefore, VLCFA-containing sphingolipids probably play a role in establishing the identity of protodermal/epidermal cells as integral constituents of the outer lateral membrane.

In Arabidopsis, the importance of cell position, rather than cell lineage, in determining cell fate is widely accepted (Stewart and Dermen, 1975; van den Berg et al., 1995; Scheres, 2001; Leyser and Day, 2009). In contrast, our analysis showed that both outermost cell lineage and outermost cell position contribute to ATML1/PDF2-mediated differentiation of protoderm/epidermis, and thus we propose that VLCFA-Cers are essential components of cell lineage- and position-dependent positional signaling.

Our analysis revealed that an unidentified candidate signal directs position-dependent post-translational regulation of ATML1 protein. It has been shown that plant subepidermal cells, which derive from asymmetric periclinal divisions of epidermal cells, develop according to their position rather than to their outermost cell lineage (Stewart and Dermen, 1975). Thus, even though the cells are directly derived from an outermost cell lineage, additional layers of regulation seem to prevent inheritance of the expression of ATML1 and PDF2 in an inner-cell position. Our assay in which ATML1 was induced by heat treatment revealed that ectopic ATML1 protein turns over more rapidly in inner cells than in outermost cells. This is consistent with our observations in 16-cell stage gATML1-EGFP;atml1-3; pdf2-1 embryos, the inner cells of which are directly derived from outermost cells. In the majority of 16-cell-stage embryos, ATML1-EGFP fluorescence was limited to the outermost cells (Fig. S3E), although EGFP fluorescence was sometimes observed in the inner cells (Fig. S3D). Considering the tight repression of transcription from the ATML1 promoter in the inner cells of 16-cell stage embryos (Takada and Jurgens, 2007), it seems that, even though ATML1-EGFP is passed on to inner cells, the protein disappears from these cells without activating the transcription of ATML1-EGFP. Taken together, we propose that the post-translational regulation of ATML1 protein observed in this study represents the position-based cell fate decision that prevents ectopic protoderm/epidermis differentiation in inner cells derived from outermost lineage cells.

Such inner cells derived from outermost lineage cells are generated through the asymmetric cell division (ACD) of outermost cells. ACD is a common mechanism for generating cell fate diversity. In Arabidopsis, BREAKING OF ASYMMETRY IN THE STOMATAL LINEAGE (BASL), which is embedded in the plasma membrane distal to the future division plane of a meristemoid mother cell, is a key regulator in the assignment of asymmetric cell fate via ACD in the stomatal cell lineage (Dong et al., 2009; Zhang et al., 2015, 2016). The polarized localization of BASL is stabilized by its phosphorylation, which is regulated by a mitogen-activated protein kinase (MAPK) signaling cascade, and, in turn, phosphorylated BASL recruits the MAPK signaling as a scaffold. Thus, the BASL-MAPK signaling feedback loop is reinforced in a polarized manner, and hence MAPK signaling asymmetrically phosphorylates SPEECHLESS, an essential TF for stomatal differentiation.

As mentioned above, we showed that the ATML1 protein itself is segregated to both the inner and outermost cells after the periclinal asymmetric division of outermost cells (Fig. S3C-E). Given that asymmetric cell fate assignment through ACD in the protodermal cell lineage is similar to that in the stomatal cell lineage, the unidentified candidate signaling component that controls post-translational regulation of ATML1 protein may be asymmetrically localized in the plasma membrane before ACD of eight-cell stage embryos. In our present work, VLCFA-Cers, which are lipidic ligand candidates of ATML1, emerge as the unidentified candidate signaling components that are asymmetrically localized in the plasma membrane, although this will remain speculative until the authentic lipid species that interact with the ATML1 protein in vivo are identified. Because of the highly redundant role of ATML1 and PDF2 in plant development and the high amino acid sequence similarity (>92%) between the START domains of ATML1 and PDF2, PDF2 is predicted to interact with the same ligand as ATML1. Loss of VLCFAs or VLCFA-containing sphingolipids results in embryonic lethality (Bach et al., 2008; Beaudoin et al., 2009; Markham et al., 2011) and severe defects in LRP development (Roudier et al., 2010; Markham et al., 2011), similar to atml1; pdf2 double mutants. These findings further support a model in which VLCFA-Cer interacts with ATML1 and PDF2.

In protodermal/epidermal cells, a small proportion of ATML1 is present in the cytoplasm (Iida et al., 2019). Thus, an interaction between ATML1 and membrane lipids may occur in the cytoplasm. In mammalian cells, several ceramide species are generated as signaling molecules by hydrolysis of sphingomyelin in the outer leaflet, and they interact with cytoplasmic target molecules via trans-bilayer movement (Stancevic and Kolesnick, 2010). In addition, the activity of sphingomyelinase (SMase) is commonly exerted in microdomains and regulated by microdomain composition (Silva et al., 2009). Our present work shows that VLCFA-sphingolipids are involved in the outer-inner polarity of protodermal/epidermal cells, suggesting that VLCFA-sphingolipids are embedded predominantly in the outer lateral membrane of protodermal/epidermal cells as microdomains similar to mammalian polarized epithelial cells. Taken together, the generation of VLCFA-Cers by an equivalent of SMase may take place in the VLCFA-sphingolipid-rich microdomains in the outer lateral membrane. In plant cells, glycosyl inositol phosphoceramide (GIPC) in the outer leaflet is a major sphingolipid of plasma membranes (Cacas et al., 2016), and complete loss of GIPC results in embryonic lethality (Tartaglio et al., 2017). Although not yet identified, a GIPC-specific phospholipase C that is predominantly located in the outer lateral membrane of the outermost cells may be a good candidate for an additional key regulator in generating a ligand for ATML1 in a position-dependent manner (Fig. S13).

Based on our finding of an outermost cell-specific protein-lipid association, we can propose a working model regarding the molecular determination of outermost and inner cells during the radial patterning process (Fig. 8). Outer lateral membrane components of the outermost cells are not passed on to either inner daughter cells through periclinal divisions or outermost cells derived from inner lineages. As a result, VLCFA-Cers generated in VLCFA-sphingolipid-rich microdomains in the outer lateral membrane by an equivalent of SMase might relay positional information along the radial axis in plants.

The START domain is also conserved in HD-ZIP class III proteins, which are key TFs in vascular development and leaf polarity, although their functional significance remains unknown (Ariel et al., 2007). Given that the association between the START domain and the homeodomain is confined to the plant kingdom (Schrick et al., 2004), such a plant-specific lipid-protein complex that mediates positional signals might represent an inevitable evolutionary strategy for sessile plants.

Columbia (Col) was used as wild-type Arabidopsis (Arabidopsis thaliana) in this study. atml1-1, atml1-3 and pdf2-1 have been described previously (Abe et al., 2003; Ogawa et al., 2015). SALK_117051, obtained from the Salk Institute Genomic Analysis Laboratory T-DNA insertion lines (http://signal.salk.edu), was backcrossed with wild type twice and named pas2-3. Primer for genotyping information is listed in Table S1.

gATML1-EGFP; atml1-3; pdf2-1 (#4), gPDF2-EGFP; atml1-3; pdf2-1 (#3), gPAS2-EGFP; pas2-3 (#16), HSP::NLS-mCherry; HSP::ATML1^WT-EGFP (#4) and HSP::NLS-mCherry; HSP::ATML1^W471L-EGFP (#1) transgenic Arabidopsis were generated in this study. The proREM1.2::YFP-REM1.2 transgenic plant has been described previously (Jarsch et al., 2014) and was obtained from T. Ott (University of Freiburg, Germany) and M. Nagano (Ritsumeikan University, Japan). The proWBC11::GFP-WBC11 transgenic plant (Luo et al., 2007) was obtained from X-Y. Chen (Chinese Academy of Sciences, China) and H. Tanaka (Meiji University, Japan). Plants were grown on soil or Murashige and Skoog (MS) solid medium supplemented with 1% sucrose at 23°C under long-day conditions (16 h light/8 h dark).

Sequences of the primer used in this study are shown in Table S2. All of the PCRs were performed using PrimeSTAR Max DNA Polymerase (Takara). To generate the mutated ATML1 (ATML1^W471), two truncated coding sequences of ATML1 with W471L mutation (ATML1m1 and ATML1m2) were amplified by PCR. The ATML1m1 fragment was amplified using the primer sets ATML1m1-F and ATML1m1-R, and the ATML1m2 fragment was amplified using the primer sets ATML1m2-F and ATML1m2-R. For the pRI201AN_ATML1^W471L, the resulting PCR products were cloned together into NdeI/SalI-digested pRI201AN vector (TaKaRa) using the In-Fusion HD Cloning Kit (TaKaRa). pRI201AN_ATML1^W471L was used as the PCR template for plasmid constructions in this study.

To construct plasmids for heat-pulse induction assay, the promoter region of HSP18.2 gene was amplified by PCR with primers HSPp-F and HSPp-R. The protein coding sequences of NLS-mCherry (Primers: NLS-mCherry-F and NLS-mCherry-R), ATML1 (Primers: ATML1-F and ATML1-R), ATML1^W471L (Primers: ATML1-F and ATML1-R) and EGFP (Primers: EGFP-F and EGFP-R) were amplified using appropriate templates by PCR. Subsequently, the PCR-amplified HSP promoter and NLS-mCherry were cloned together into PstI/SalI-digested pRI201AN vector using In-Fusion HD Cloning Kit to generate pRI201AN_HSP::NLS-mCherry::HSP ter. The PCR-amplified HSP promoter, ATML1 and EGFP fragments were cloned together into the PstI/SalI-digested pRI201AN vector to generate pRI201AN_HSP::ATML1-EGFP::HSP ter. The PCR-amplified HSP promoter, ATML1^W471L, and EGFP fragments were cloned together into PstI/SalI-digested pRI201AN vector to generate pRI201AN_HSP::ATML1^W471L-EGFP::HSP ter. Finally, the DNA fragments of HSP::ATML1-EGFP::HSP ter and HSP::ATML1^W471L-EGFP::HSP ter were amplified by PCR with primers HSP_ATML1_F and HSP_ATML1_R, and cloned into SmaI site of HSP::NLS-mCherry::HSP ter to generate HSP::NLS-mCherry; HSP::ATML1^WT-EGFP and HSP::NLS-mCherry; HSP::ATML1^W471L-EGFP, respectively.

To generate the gATML1-EGFP plasmid, the DNA fragment (5′-CACCAATAACTTCGTATAGCATACATTATACGAAGTTATGGATCCCGAATTCATAACTTCGTATAGCATACATTATACGAAGTTATT-3′) was inserted into pENTR/D-TOPO Vector (Thermo Fisher Scientific) according to the manufacturer's instructions (pENTR/D_loxP_BamHI-EcoRI_loxP). The sequences of EGFP (primers EGFP2-F and EGFP2-R) and the 3′ UTR sequence of ATML1 (primers ATML1ter-F and ATML1ter-R) were then respectively amplified by PCR. The resulting PCR products were cloned together into EcoRI site of the pENTR/D_loxP_BamHI-EcoRI_loxP vector using the In-Fusion HD Cloning Kit to generate pENTR/D_loxP_BamHI-EcoRI-EGFP-ATML1 ter-loxP. Next, promoter and the protein-coding sequence of ATML1 were amplified by PCR with primers gATML1-F and gATML1-R, and cloned into BamHI/EcoRI-digested pENTR/D_loxP_BamHI-EcoRI-EGFP-ATML1 ter-loxP (pENTR/D_loxP-gATML1-EGFP-loxP). Finally, for gATML1-EGFP, we recombined pENTR/D_loxP-gATML1-EGFP-loxP into the binary expression vector (pHGW.0) using Gateway LR Clonase II enzyme mix (Thermo Fisher Scientific).

To construct gPDF2-EGFP, we first respectively amplified the sequences of EGFP (primers EGFP3-F and EGFP3-R) and 3′ UTR sequence of PDF2 (primers PDF2ter-F and PDF2ter-R) by PCR. The resulting PCR products were cloned together into BamHI/EcoRI-digested pENTR/D_loxP_BamHI-EcoRI_loxP vector using the In-Fusion HD Cloning Kit to generate pENTR/D_loxP_BamHI-EGFP-PDF2 ter-EcoRI-loxP. Next, promoter and the protein-coding sequence of PDF2 were amplified by PCR with primers gPDF2-F and gPDF2-R, and cloned into the BamHI site of pENTR/D_loxP_BamHI-EGFP-PDF2 ter-EcoRI-loxP (pENTR/D_loxP-gPDF2-EGFP-loxP). Finally, for gPDF2-EGFP, we recombined pENTR/D_loxP-gPDF2-EGFP-loxP into a binary expression vector (pHGW.0) using the Gateway LR Clonase II enzyme mix.

To construct gPAS2-EGFP, we first amplified the sequence of EGFP by PCR (primers loxEGFP-F and loxEGFP-R) and cloned into the PstI/EcoRI-digested pRI201AN vector using the In-Fusion HD Cloning Kit (pRI201AN_loxP-EGFP-loxP). Next, the 3′ UTR sequence of PAS2 was amplified by PCR with primers PAS2ter-F and PAS2ter-R, and cloned into the EcoRI site of pRI201AN_loxP-EGFP-loxP to generate pRI201AN_loxP-EGFP-PAS2 ter-loxP. Finally, for gPAS2-EGFP, the promoter and protein-coding sequence of PAS2 were amplified by PCR (primers gPAS2-F and gPAS2-R), and cloned into the BamHI site of pRI201AN_loxP-EGFP-PAS2 ter-loxP.

To construct vectors for the protein expression, DNA fragments corresponding to the START domain of ATML1 and ATML1^W471L were amplified by PCR (primers START-F and START-R). The resulting PCR fragments were cloned into the NdeI/XbaI-digested pCold GST DNA (TaKaRa) to generate pCOLD_GST-START and pCOLD_GST-START^W471L, respectively.

The constructs described above in binary vectors were introduced into Agrobacterium tumefaciens strain GV3101 and transformed into Arabidopsis plants by the floral-dip procedure (Clough and Bent, 1998).

For induction of HSP18.2 promoter activity, Arabidopsis seedlings were submerged into 1/2 MS liquid medium and incubated in a water bath for 3 h at 35°C. For RT-PCR analysis, total RNA was extracted using TRIzol reagent (Thermo Fisher Scientific) and was treated with RNase-free DNase (Thermo Fisher Scientific) according to the manufacturer's instructions. Total RNA (1 μg) was reverse transcribed with oligo dT primer according to the protocols of the Superscript II (Thermo Fisher Scientific). The primer sequence and PCR conditions are provided in Table S3.

Both FB1 (Sigma-Aldrich) and cafenstrole (FUJIFILM Wako Pure Chemical Corporation) were dissolved in dimethyl sulfoxide (DMSO) and added to media so that the solvent concentration did not exceed 0.1% (v/v). For long-term treatment, seedlings were directly germinated on MS solid media supplemented with 0.5 μM FB1 or 0.3 μM cafenstrole. For FB1 short-term treatment, 7-day-old seedlings were transferred to 1/2 MS liquid media supplemented with 3 μM FB1 and incubated for 24 h. In each experiment, mock treatment [0.1% (v/v) DMSO treatment] was performed as a control experiment.

Confocal microscopic analysis was performed using an LSM 510 Meta Confocal laser-scanning microscope (Carl Zeiss) or a C2 confocal Microscope (Nikon). All microscopic analyses were performed multiple times, and all repeats were successful.

Total RNA was extracted from 5-day-old seedlings using TRIzol reagent and was treated with RNase-free DNase. Total RNA (1 μg) was reverse transcribed with oligo dT primer according to the protocols of the Superscript II.

RT-qPCR was performed with the Light Cycler 480 II (Roche) using Light Cycler 480 Probe Master (Roche). Three biological replicates were used for the quantification. The mean of housekeeping gene PP2A (AT1G13320) was used as an internal control to normalize the variability in expression levels. The primer sequence and probe list are provided in Table S4.

Sphingolipid contents in lyophilized Arabidopsis tissues were determined using LC-MS/MS with a targeted monitoring method as described previously (Ishikawa et al., 2018). Escherichia coli Rosetta 2 strain (Novagen) expressing pCOLD_GST-START or pCOLD_GST-START^W471L constructs were grown in LB (Luria-Bertani) liquid medium till an OD600 of 0.4-0.6, and then transferred to an ice-cold water bath for 10 min. After 10 min incubation, cell cultures were incubated at 15°C for 30 min, and then the expression of GST tagged recombinant proteins was induced by 1 mM isopropyl-thiogalactoside (IPTG). After induction, cells were cultivated for 24 h at 15°C. Subsequently, the cells were harvested by centrifugation and the cell-pellet was resuspended in 10 ml of Bugbuster master mix and incubated at room temperature for 10 min. The cell suspension was centrifuged and the supernatant, which contains soluble GST-tagged recombinant proteins, was collected. GST-tagged recombinant proteins were purified using glutathione sepharose 4B (GE Healthcare) according to the manufacturer's instructions.

For the sphingolipids-protein interaction assay, sphingolipid-spotted membrane was originally prepared. C16 Ceramide (d18:1/16:0), C24 Ceramide (d18:1/24:0), C16 Ceramide-1-Phosphate (d18:1/16:0) and C24 Ceramide-1-Phosphate (d18:1/24:0) were purchased from Avanti Polar Lipids. Glucosylceramide from rice (Glucosylceramide mix) was purchased from Nagara Science. These lipids were dissolved in methanol/chloroform (1:1) to a concentration of 100 μM. GIPC fractions were prepared from Arabidopsis shoots by a method using isopropanol/hexane/water extraction, followed by fractionation with weakly anion exchange chromatography as described previously (Ishikawa et al., 2016). The purity of the GIPC fraction was determined by LC-MS/MS, in which no other sphingolipid classes were detected. The fraction was finally dissolved in 2-propanol/water/hexane (55:20:5) to a concentration of 100 μM. Aliquots of each lipid solution (2 μl) were directly spotted onto nitrocellulose membrane (Amersham Protran Supported 0.45 μm). For lipid-protein interaction assays other than those above, Sphingo Strips and PIP Strips were obtained from Echelon Biosciences. The lipid overlay assay was performed as described previously (Dowler et al., 2002). The membranes were blocked for 1 h at room temperature in blocking buffer [2 mg/ml fatty-acid-free BSA, 150 mM NaCl, 50 mM Tris (pH 7.5), 0.1% (v/v) Tween 20]. The membranes were then incubated for 2 h at room temperature in the presence of 1 μg/ml purified GST tagged recombinant proteins. The membranes were probed with a polyclonal anti-GST antibody [MBL (PM013); 1:1000 in blocking buffer] for 1 h at room temperature. The secondary antibody, anti-rabbit IgG HRP-linked whole Ab donkey [GE Healthcare (NA934); 1:5000 in TBST], was incubated with the membrane for 1 h at room temperature. Bound antibody was visualized with the Ez West Lumi One (ATTO). The signals were collected using the Image Quant LAS 4000 mini (GE Healthcare). The mean of signal intensity was measured by using ImageJ software.

We thank A. Watanabe-Taneda and A. Imai for experimental assistance.