Sterols regulate endocytic pathways during flg22-induced defense responses in Arabidopsis

Plants possess a multilayered defense system of innate immunity that confers resistance against many pathogens. Plant cells have a variety of immune receptors, also known as pattern recognition receptors (PRRs), that recognize conserved microbial- or pathogen-associated molecular patterns (PAMPs) (Dodds and Rathjen, 2010; Monaghan and Zipfel, 2012; Felix et al., 1999), such as fungal chitin, and the bacterial factor elongation factor EF-Tu (Kunze et al., 2004), flagellin (Robatzek et al., 2006) and lipopolysaccharides (Mulder et al., 2006; Radutoiu et al., 2007). The Arabidopsis thaliana receptor kinase FLAGELLIN SENSING 2 (FLS2) is a well-known PRR that perceives a conserved 22 amino acid domain in the N terminus of flagellin (flg22) (Mulder et al., 2006; Radutoiu et al., 2007). Upon perception of flg22, FLS2 physically associates with another leucine-rich repeat receptor-like kinase (LRR-RLK), BAK1 (BRASSINOSTEROID-INSENSITIVE 1-ASSOCIATED KINASE 1) (Heese et al., 2007; Roux et al., 2011), and is phosphorylated by BIK1 (BOTRYTIS-INDUCED KINASE 1) (Segonzac and Zipfel, 2011). Activation of FLS2 triggers a series of downstream immune responses (Asai et al., 2002; Boudsocq et al., 2010; Henry et al., 2013), including the production of reactive oxygen species (ROS), induction of defense genes, deposition of secondary compounds, such as callose, and accumulation of defense hormones (Grossie et al., 2009; Gómez-Gómez and Boller, 2000; Boudsocq et al., 2010).

Numerous reports support the concept that endocytosis of receptor kinases physically terminates signaling through degradation of receptors, sustains signaling through recycling or relays signals inside the cell through the formation of signaling endosomes (Du et al., 2013). The recycling/signaling and the degradative fates preferentially associate with different endocytic routes (Sigismund et al., 2008). In animal cells, multiple pathways of endocytosis have been identified and classified, including clathrin-dependent and clathrin-independent pathways (Doherty and McMahon, 2009). Similarly, two endocytic pathways have been found in plant cells: the clathrin-mediated and the sterol/raft-associated endocytic pathways (Fan et al., 2015). The clathrin-mediated endocytic pathway (CME) is the major and best-studied endocytic pathway in plant cells (Ortiz-Morea et al., 2016).

The sterol/raft-associated endocytic pathway has also been described and may involve specific membrane microdomain proteins such as flotillin 1 (Flot1) (Li et al., 2012). Membrane microdomains are thought to temporally and spatially organize proteins and lipids into dynamic signaling complexes (Cacas et al., 2012). In animal and yeast cells, sterol-rich domains may function as sorting platforms for proteins that function in signal transduction, pathogen entry, secretion and endocytosis (Simons and Toomre, 2000). Sterol-associated endocytic pathways have been described in plant cells (Men et al., 2008), although the mechanisms of this pathway in plant immunity have yet to be investigated.

Here, we describe a series of investigations into the mechanism of FLS2 endocytosis, to explore the link between sterols and FLS2 dynamics in response to the plant immunity signal flg22. Phenotypic analyses showed that sterols can affect the flg22-induced immune response. Variable-angle total internal reflection fluorescence microscopy (VA-TIRFM) in combination with single-particle tracking (Li et al., 2011) showed that FLS2 is mobile and heterogeneously distributed at the plasma membrane, and that sterols affected this distribution. Furthermore, we demonstrate that the sterol methyltransferase 1 (smt1) mutation does not change the homo-oligomeric state of FLS2 but does affect FLS2 cluster formation. In addition, endocytosis of FLS2 is impaired in smt1 mutants. These multifaceted approaches support the hypothesis that sterol-associated endocytic pathways are essential for the flg22-induced dynamics of FLS2 and plant defense.

Plant sterols are key components of membrane microdomains and are connected with defense responses (Kopischke et al., 2013). Interestingly, we found that treatment with the sterol-extracting reagent methyl-β-cyclodextrin (MβCD) increased flg22-induced ROS production (Fig. S1). To further examine the role of sterols in plant immune pathways, we analyzed the flg22-induced immune response in the sterol biosynthesis mutant sterol methyltransferase 1 (smt1) (Diener et al., 2000). SMT1 is a C-24 methyltransferase that catalyzes the first step of sterol biosynthesis, the conversion of cycloartenol to 24-methylenecycloartenolin, and the smt1 mutant has defects in sterol composition (Diener et al., 2000; Clouse, 2002). As shown in Fig. 1A,B, ROS rapidly accumulated in wild-type plants treated with flg22. However, in the smt1 mutants, flg22-induced ROS production was significantly higher (up to 40%) compared with the wild type.

To confirm whether the increased ROS production was due to increased expression of FLS2 in the smt1 mutants, we performed quantitative reverse transcription PCR (qRT-PCR) to measure FLS2 expression. We observed no significant difference in FLS2 mRNA levels between the wild-type and the smt1 mutant plants (Fig. 1C). Furthermore, immunoblot analysis showed that the FLS2 protein levels were comparable in the wild type and the smt1 mutant lines (Fig. 1D), indicating that the steady-state levels of the FLS2 receptor likely cannot account for the observed increase in flg22-induced ROS production in the smt1 mutants, and that sterols affect the flg22-induced ROS burst.

Therefore, we subsequently investigated the role of sterols in flg22-induced plant immunity. By investigating flg22-induced calcium influx, one of the earliest signaling events that occurs within 2-5 min after PAMP perception (Smith et al., 2014), we found that the smt1 mutants exhibited a rapid and strong increase in [Ca²⁺]_cyt when compared with the wild-type plants (Fig. 1E,F), suggesting that, compared with wild type, the smt1 mutants were more sensitive to the flg22 treatment.

MAP kinases (MAPKs) also play important roles in plant defenses and we next examined flg22-induced MAPK activity in the smt1 mutants. CYP81 and WRKY40 are MAPK-dependent marker genes (Mao et al., 2011; Boudsocq et al., 2010). We found that the increases of CYP81 and WRKY40 transcript levels induced by treatment with flg22 were significantly lower in the smt1 mutants when compared with the wild-type plants 30 min after treatment (Fig. 1G). We further monitored MAPK activation by measuring MPK phosphorylation in response to flg22 treatment. As shown in Fig. S2, the phosphorylation levels of MPK3, MPK4 and MPK6 were significantly lower in the smt1 mutants than in wild type, consistent with the qRT-PCR results measuring CYP81 and WRKY40 expression (Fig. 1G).

Other studies have examined the effect of brassinosteroids on smt1 plants (Diener et al., 2000). The root growth of smt1 mutants is extremely sensitive to epibrassinolide (eBL) treatment (Diener et al., 2000) and eBL can suppress flg22-mediated immunity responses (Belkhadir et al., 2012; Albrecht et al., 2012). However, in smt1 plants, reduced endogenous brassinolide biosynthesis caused by introducing the brassinosteroid-deficient mutant deetiolated2 (det2) into the smt1 background did not rescue the poor growth of smt1 roots (Diener et al., 2000; Carland et al., 2010), suggesting that the phenotypic abnormalities of smt1 are not due to the accumulation of brassinolides. Therefore, we inferred that the immune defects after flg22 treatment resulted mainly from the altered sterol biosynthesis in smt1 mutants.

To further investigate the role of sterols in flg22-induced defenses, we examined callose deposition, another important marker of PAMP-mediated immunity. The flg22-induced callose deposition was significantly impaired in the smt1 mutants when compared with the wild type 24 h post-infiltration of active flg22 (Fig. 1H and Fig. S3), although no difference was detected in the plants infiltrated with H₂O (Fig. 1H and Fig. S3). These analyses of the smt1 mutants indicate that sterols play a profound role in flg22-induced plant immunity. Consistent with this, accumulating evidence shows that plant immunity is achieved by the formation of nanometer-scale membrane domains (termed microdomains), which act as signaling hubs (Keinath et al., 2010; Stanislas et al., 2009) and sterols are the main lipid components of these membrane microdomains.

Membrane-localized receptors sense environmental cues that trigger a physiological response (Li et al., 2013). The flg22 receptor FLS2 is a well-studied leucine-rich repeat receptor in plant immunity, and is used for studies of receptor-mediated immunity in plants (Yeh et al., 2015). Although previous studies have addressed flg22-induced defense responses (Ben Khaled et al., 2015), the dynamic behavior of FLS2 in the plasma membrane remains largely unknown.

Precise analyses with high spatial and temporal accuracy are required to characterize the dynamics of membrane receptors; e.g. VA-TIRFM and single-particle tracking of fluorescent proteins can be used to image membrane proteins and measure their dynamic movements. To probe the dynamics of FLS2 in the membrane, we therefore generated transgenic Arabidopsis plants expressing FLS2 fused with GFP under the control of the native FLS2 promoter in the fls2 mutant. Immunoblot analysis showed that the FLS2-GFP fusion was expressed (Fig. S4A) and FLS2-GFP fully rescued the ROS production (Fig. S4B) and callose deposition defects of the fls2 mutant (Fig. S4C), indicating that FLS2-GFP is functional. We next used VA-TIRFM to image FLS2-GFP in living leaf epidermal cells. The FLS2-GFP fluorescent signal appeared as discrete spots at the plasma membrane (Fig. 2A,B) and we applied VA-TIRFM to analyze the dynamics of these spots (Wang et al., 2018). We found that numerous FLS2-GFP fluorescent spots disappeared from or came into the focal plane during live-cell imaging (Fig. 2C,D and Movie 1), indicating exocytosis or endocytosis of plasma membrane components. Single-particle tracking analysis of the time series of images revealed that FLS2 is highly mobile and heterogeneously distributed at the plasma membrane.

Sterols are integral components of the cell membrane and affect the lateral mobility of plasma membrane proteins (Saka et al., 2014). To study the effect of sterols on the diffusion of FLS2-GFP, we measured the range of motion of FLS2-GFP in the smt1 mutants. Under control conditions, the motion range of FLS2 showed a bimodal distribution: long-distance motion (67.8%, 3.05±0.41 μm, the Gaussian peak value) and short-distance motion (32.2%, 1.73±0.13 μm, the Gaussian peak value) (Fig. 2E,I). Upon flg22 treatment, the percentage of FLS2-GFP exhibiting short-distance motion decreased to 20.6% and those exhibiting long-distance motion increased to 79.4% compared with the control (Fig. 2E,I and Movie 2), indicating that flg22 treatment activated long-distance FLS2-GFP movement.

In the smt1 mutants carrying FLS2-GFP, long-distance and short-distance motion accounted for 63.2% and 36.8% (Fig. 2F,I and Movie 3), respectively, indicating that the sterol defect in the mutants slightly affected the range of FLS2-GFP motion. Moreover, under flg22 treatment, the percentage of long-distance and short-distance motion of FLS2-GFP accounted for 73.1% and 26.9% in smt1 mutants, which was significantly different from the distribution in wild-type Col-0, implying that the smt1 mutation affects the motion range of flg22-activated FLS2-GFP (Fig. 2F,I). This finding was confirmed by treatment with MβCD, which decreased the percentage of long-distance motion of flg22-activated FLS2 (Fig. S5A,B and Movie 4), suggesting that reduction of the amounts of sterols significantly affected the range of motion of flg22-activated FLS2-GFP.

We further plotted the distribution of diffusion coefficients of FLS2-GFP on histograms and fitted the histogram using the Gaussian function, in which we defined the Gaussian peaks (noted as Ĝ) as the characteristic values. In the control, the Ĝ values were 7.9×10⁻³ μm²/s (s.e. 7.72 to 8.18×10⁻³ μm²/s) (Fig. 2G,J). In response to the flg22 treatment, the Ĝ values were 1.48×10⁻² μm²/s (s.e. 1.44 to 1.51×10⁻² μm²/s) (Fig. 2G,J and Movie 2), representing a marked increase in diffusion compared with the control plant. To further confirm that the effect was specific to flg22, we used flg22Δ2, a flg22-derived peptide lacking agonist activity (Danna et al., 2011). When the seedlings were treated with flg22Δ2, no significant effect was observed on the range of motion and diffusion coefficients of FLS2-GFP when compared with the control untreated plant (Fig. 2I,J). In the smt1 mutants, there was a slight but not significant increase in the diffusion coefficient (from 7.9×10⁻³ μm²/s to 8.85×10⁻³ μm²/s) of FLS2-GFP when compared with that in Col-0 without flg22 treatment (Fig. 2H,J and Movie 3). Notably, the diffusion coefficient was 1.07×10⁻² μm²/s in smt1 mutants under flg22 treatment, significantly lower than that in Col-0 under flg22 treatment (Fig. 2H,). A similar result was observed when the plants were treated with MβCD (Fig. S5C,D and Movie 4), supporting the idea that depletion of sterols significantly affects the dynamic behavior of flg22-activated FLS2.

A previous study showed that flg22 perception initiates the innate immune response (Li, et al., 2014). In addition, sterol-rich domains may modulate the dynamics and activity of some membrane proteins (Wang et al., 2015; Xue et al., 2018), which are involved in PAMP-induced immunity signaling (Keinath et al., 2010). In the present investigation, we showed that flg22-activated FLS2-GFP spots move faster and diffuse into wider regions in comparison with inactivated FLS2-GFP, indicating that the dynamic behaviors of FLS2 were closely related to flg22-induced signaling. Moreover, disruption of sterols by MβCD treatment or in smt1 mutants significantly affected flg22-induced plant immunity (Fig. 1) and the diffusion of FLS2-GFP (Fig. 2I,J). Therefore, we concluded that sterols are involved in the flg22-induced immune responses, acting by regulating FLS2 motions.

Previous studies have shown that the binding of signaling molecules to activated receptors resulted in receptor endocytosis (Schlessinger, 2002). Single-particle tracking provides a powerful method with which to study membrane protein state and local environment, which can constrain protein dwell time (Flores-Otero et al., 2014; Hao et al., 2014). As flg22-activated FLS2 receptors can internalize into endosomes, we attempted to investigate whether sterols affect the dwell time of FLS2 at the plasma membrane. Therefore, we used VA-TIRFM to investigate the change in dwell time of FLS2 particles in response to flg22 treatment. When the frequency distribution of the FLS2-GFP dwell times under the different treatments was fitted to an exponential function, we found that flg22 treatment led to significantly shorter dwell times (τ=1.24 s) than that in the control seedlings (τ=2.15 s) (Fig. 3A-D,I). However, MβCD treatment (τ=2.73 s) resulted in longer dwell times than that the control seedlings (Fig. 3E,F,I), whereas the dwell times of FLS2 became longer in the cells treated with MβCD together with flg22 (τ=1.91 s) when compared with flg22 treatment (Fig. 3G,H,I). From the different dwell times of FLS2 at the plasma membrane in the various treatments, we speculated that the internalization of FLS2 into endosomes is a highly efficient way to control FLS2 activity and FLS2 endocytosis may occur through more than one endocytic pathway.

The lateral mobility of many membrane receptor proteins depends on their oligomerization status (Low-Nam et al., 2011). Activation of membrane receptor proteins generally involves ligand-induced homo/heterodimerization or hetero-oligomerization in membrane microdomains and leads to signal transduction (Schlessinger, 2002; Corriden et al., 2014). In a previous study, fluorescence resonance energy transfer (FRET) and bimolecular fluorescence complementation (BiFC) assays showed that FLS2 does not undergo homodimerization either before or after flg22 treatment in Arabidopsis protoplasts (Ali et al., 2007), consistent with protein crystal structure studies (Sun et al., 2013). In the present study, by analyzing sequential images, we found that two FLS2 fluorescence spots diffused laterally along the membrane, collided and then fused together (Fig. 4A). By examining photobleaching steps, which reflect the oligomerization status of the target proteins (Ulbrich and Isacoff, 2007; Das et al., 2007), we found that most of the FLS2-GFP molecules exhibited one-step bleaching, and a few exhibited two-step bleaching (Fig. 4B,C) indicating that FLS2 at the plasma membrane of living cells exists in multiple oligomeric forms, including monomers and dimers.

To clarify whether FLS2 exists as dimers, we carried out luciferase complementation assays in Nicotiana benthaminana plants, with FLS2-Nluc and the non-interacting protein Cluc-CONSTITUTIVE PATHOGEN RESPONSE 5 (CPR5) as a negative control and FLS2-Cluc plus BIK1-Nluc as a positive control. The results showed that FLS2 can interact with itself and BIK1 but not with CPR5 in the absence or in presence of flg22 (Fig. 4D). To further verify the dimerization of FLS2, we performed co-immunoprecipitation assays in Arabidopsis protoplasts. For this purpose, FLS2-HA/FLS2-FLAG and FLS2-HA/BAK1-FLAG were co-transfected into Arabidopsis protoplasts. As shown in Fig. 4E, FLS2-HA was detected in the FLS2-FLAG immune complex using anti-HA antibodies both in the absence and presence of flg22 (Fig. 4E), demonstrating that FLS2 can form a homo-dimer either in the inactive or active state.

We next addressed whether FLS2 was predominantly present as a monomer or dimer. Using progressive idealization and filtering (PIF) algorithms, we found that the ratios of the monomer versus the dimer of FLS2-GFP in an inactive state (0.86:0.14) were similar to the ratios in the active state (0.84:0.16) (Fig. 4F), indicating that monomers are the predominant species. In addition, dimers were detected before and after flg22 treatment. Moreover, with the MβCD treatment, the ratio of the monomer versus the dimer was comparable with that in the control (Fig. 4F), suggesting that sterols do not affect the homo-oligomeric state of FLS2.

Receptor-ligand binding can induce conformational changes of the receptor ectodomain, which result in receptor dimerization and receptor oligomerization or clustering (Clayton et al., 2005). With VA-TIRFM, we observed that most of the FLS2-GFP existed in low-oligomeric spots with an average size of 2.43×2.43±0.92 pixels and a fluorescence intensity of 6968.4±697 counts/pixel in the control plants (Fig. 5A-C). Under flg22 treatment, the FLS2-GFP spots became larger and their fluorescence intensity became brighter when compared with those in the control plant (Fig. 5A). We found that the large spots exhibited exponential decay without discrete steps (Fig. 5D), suggesting that each large spot was a cluster of several FLS2-GFP molecules. The flg22-induced clusters exhibited an average size of 4.12×4.12±1.77 pixels and a fluorescence intensity of 12,254.5±1505 counts/pixel after 30 min (Fig. 5B,C). When treated with MβCD coupled to flg22, the average size and fluorescence intensity of FLS2-GFP spots significantly decreased to 2.83×2.83±0.96 pixels and 8649.5±1291 counts/pixel (Fig. 5B,C), suggesting that the MβCD treatment resulted in a reduction of flg22-induced FLS2-GFP clusters at the plasma membrane (Fig. S6).

Consistent with previous studies (Gao et al., 2015; Wang et al., 2015), Zipfel and colleagues found that cluster formation relies on the integrity of the plasma membrane, as MβCD-induced sterol depletion of the plasma membrane destroys clustering (Bücherl et al., 2017). Furthermore, clustering is required for kinase activity, and abnormal protein complex organization can affect the internalization of membrane proteins (de Heus et al., 2013). Therefore, our findings suggest that sterols are involved in modulating FLS2 signal transduction by adjusting the clustering of FLS2 induced by flg22 at the plasma membrane.

Clustering can promote endocytosis of plasma membrane proteins and this regulates signaling of the receptor (Hofman et al., 2010). FLS2-GFP translocation from the plasma membrane to endosomes is mainly mediated by clathrin (Ben Khaled et al., 2015; Mbengue et al., 2016). We applied fluorescence correlation spectroscopy (FCS) to precisely measure the density of FLS2 molecules under different conditions. FCS is a powerful technique for analyzing particle concentrations, mobilities and diffusion at the single-particle level, and has been successfully applied in plants (Li et al., 2011). Our FCS analyses showed that the density of FLS2 molecules at the plasma membrane significantly increased after treatment with Tyrphostin A23 (TyrA23), which is an inhibitor of the clathrin-mediated endocytosis pathway (Fig. 6A). Similar results were found in equivalent experiments using immunoblot assays (Fig. 6B). In addition, using single particle tracking analysis, we found that the flg22-induced diffusion of FLS2-GFP was significantly impaired in the clathrin light chain2 (clc2) and clc3 double mutants (clc2-1 clc3-1) or using TyrA23 treatment when compared with the control plant (Fig. S7), supporting the notion that clathrin is involved in the internalization of FLS2, as reported previously (Mbengue et al., 2016; Ortiz-Morea et al., 2016).

Moreover, our experiments show that TyrA23 treatment significantly blocked the constitutive endocytosis of FLS2-GFP accumulated in brefeldin A (BFA) compartments (Fig. 6C), in seedlings treated with BFA. BFA is an inhibitor of vesicle recycling in plants, acting through inhibition of ADP ribosylation factor-guanine exchange factors (Grebe et al., 2003; Naramoto et al., 2010). Quantitative analyses revealed the internalization of flg22-induced FLS2-GFP was not fully blocked (reduced by about 40%) after TyrA23 pretreatment (Fig. 6D,E), which was consistent with the results in the clc2-1 clc3-1 mutants (Fig. S8). As a control, we analyzed the endocytosis of FLS2-GFP using Tyrphostin A51 (TyrA51), the inactive analog of TyrA23 (Wang et al., 2015), and found that TyrA51 had no significant effects on the internalization of FLS2-GFP (Fig. 6C-E). In addition, we found when co-expressing FLS2-GFP with clathrin light chain (CLC)-mCherry or with VHA-a-RFP [the trans-Golgi network (TGN) markers], FLS2-GFP showed only partial colocalization with CLC-mCherry (17.0%) or VHA-a1-mCherry (14.7%) after 30 min of treatment with flg22 (Fig. 7A). Therefore, these results provided strong evidence that more than one endocytic pathway acts to coordinate the endocytosis of FLS2.

Previous reports showed that sterol-enriched domains may play a role in pathogen-associated molecular pattern-induced signaling and affect the endocytosis of plasma membrane proteins in plants (Keinath et al., 2010; Yadeta et al., 2013; Malinsky et al., 2013). In plants, the membrane microdomain-associated endocytic pathway has also been described, and includes microdomain-localized proteins such as flotillins (Li et al., 2012). Yu et al. showed that flg22 can increased the endocytosis of GFP-Flot1 (Yu et al., 2017). Surprisingly, we found that FLS2-GFP showed a significant colocalization with AtFlot1-mCherry (marker of sterol-rich domains) endosomes (44.9%) (Fig. 7A). Given the link between membrane microdomains and endocytosis, this finding indicates that sterols may be involved in FLS2 internalization.

To test this hypothesis, we analyzed the degree of colocalization of FLS2-GFP with AtFlot1-mCherry and CLC-mCherry by calculating protein proximity indexes (PPI) (Fig. 7B,C) (Zinchuk et al., 2011). We measured the PPI values before and after flg22 treatment. The results showed that the mean FLS2-to-Flot1 PPI showed a highly significant increase (from 27.2% to 44.8%) in response to flg22. The FLS2-to-CLC PPI also increased significantly (from 48.3% to 60.5%) after 30 min of flg22 treatment (Fig. 7D), confirming that microdomains are recruited for flg22-dependent FLS2 endosomes.

To test whether sterols affect the internalization of proteins, we applied BFA to inhibit endocytic recycling (Naramoto et al., 2010). FLS2-GFP did not accumulate in BFA compartments in cells treated with MβCD and BFA when compared with BFA alone (Fig. S9). Furthermore, the number of FLS2-GFP endosomes in the smt1 mutants was notably lower than those in the control after 30 min of flg22 treatment, and this trend became more pronounced after 60 min of flg22 treatment (Fig. 7E,F), which was similar to the results with the MβCD treatment (Fig. S10A and B), indicating the existence of a close relationship between FLS2 localization and sterols. To exclude potential side effects of MβCD, we compensated for the loss of plasma membrane cholesterol by adding MβCD-cholesterol complexes (10 mM) to seedlings following MβCD treatment (Chadda et al., 2007). We found that the internalization of FLS2 recovered after sterol supplementation (Figs S9 and S10A,B), indicating the disturbance of FLS2 internalization resulted largely from the depletion of cholesterol, not from potential side effects of MβCD.

Using the relative intensity of cytoplasm/plasma membrane as a proxy for the internalization rate, we found that the ratio in the smt1 mutants was reduced to 17% after flg22 treatment, which was significantly lower than that in the wild-type plant (27.8%) (Fig. 7G). Similar results were also found in the MβCD-pretreated plants (Fig. S10C), suggesting that FLS2 internalization was impaired in the smt1 mutants.

In the present study, we precisely measured the density of FLS2-GFP on the membrane by FCS. The density of FLS2-GFP on plasma membranes was 11.72 molecules mm⁻² before flg22 treatment, and 8.93 molecules mm⁻² after flg22 treatment (Fig. S11A). After treatment with MβCD or MβCD coupled to flg22, the abundance of FLS2 molecules at the plasma membrane significantly changed compared with controls (12.87 molecules mm⁻² for MβCD and 10.47 molecules mm⁻² for MβCD coupled with flg22) (Fig. S11A), indicating that the internalization of FLS2-GFP, especially the flg22-induced internalization, was partly disrupted by MβCD. Western blotting demonstrated that the smt1 mutation can significantly reduce flg22-induced degradation of FLS2 protein compared with that in the wild type (Fig. 7H), and similar results were obtained under MβCD treatment (Fig. S11B). Sterols are associated with endocytosis (Stanislas et al., 2014; Ferrer et al., 2017). For example, Hao et al. revealed that sterol extraction by treatment with MβCD significantly disrupted NaCl-induced GFP-RbohD internalization (Hao et al., 2014). Furthermore, the sterol biosynthesis mutants cyclopropylsterol isomerase1-1 (cpi1-1) and sterol methyltransferase 1 (smt1) impaired endocytosis of PIN2 protein (Men et al., 2008; Willemsen et al., 2003). Taken together, these results further support the hypothesis that sterols coordinate the internalization of FLS2.

Endocytosis plays key roles in regulating signaling initiation at the plasma membrane, where plants can sense the external stimuli and signaling through cell surface receptor-like kinases (RLKs) (Barbieri et al., 2016; Irani and Russinova, 2009). Some RLKs (including BRI1, FLS2, EFR and PEPR1/2) have been reported to undertake receptor-mediated endocytosis following ligand perception (Irani et al., 2012; Mbengue et al., 2016; Ortiz-Morea et al., 2016). More importantly, several lines of evidence support the idea that the sterol-associated endocytosis is involved in RLK stability and endocytosis. For example, sterols are essential for the activation of BRI1 signaling and BRI1 internalization (Wang et al., 2015). Given that sterols are involved in the internalization and signal transduction of FLS2 after sensing flg22, we speculated that sterols exert a profound role in regulating RLK endocytosis, which controls receptor availability on the plasma membrane and fine-tunes the delivery of signals to diverse downstream effectors, ultimately regulating the cellular response to different types of signals.

All mutants and transgenic lines used were in the Arabidopsis thaliana Col-0 background. Arabidopsis seeds were surface sterilized for 30 s in 85% ethanol/H₂O₂ and vernalized at 4°C for 1 day in the dark. Seedlings were then grown on half-strength Murashige and Skoog (MS) medium. The plants were grown at 22°C under fluorescent light at 70% relative humidity with 16 h light/8 h dark photoperiods, unless otherwise specified. The following mutant and transgenic Arabidopsis lines have been described previously: fls2, smt1, clc2-1clc3-1, clathrin light chain (CLC)-mCherry, Flot 1-mCherry and VHA-a1-RFP (Gómez-Gómez and Boller, 2000; Diener et al., 2000; Wang et al., 2013; Li et al., 2012; Mbengue et al., 2016). Dual-color lines expressing FLS2-GFP with CLC-mCherry, Flot1-mCherry and VHA-a-RFP were generated by crossing FLS2-GFP with CLC-mCherry, Flot1-mCherry and VHA-a-RFP, respectively.

To generate the FLS2 transgenic plants, a native FLS2 promoter of 988 bp in length and the FLS2 gene double myc-tag fusion were PCR amplified and cloned into pCAMBIA2300. The FLS2 promoter sequence was PCR amplified using the primers 5′-GGAATTCGAAGTTGTGAATTGTGATCATAGGA-3′ (forward) and 5′-GGGGTACCGGTTTAGACTTTAGAAGAGTT-3′ (reverse) carrying the EcoRI and KpnI sites. The coding sequence for FLS2 (At5g46330) was PCR amplified using the primers 5′-GGGGTACCATGAAGTTACTCTCAAAGACCTTTTTG-3′ (forward) and 5′-ACGCGTCGACAACTTCTCGATCCTCGTTACGATCT-3′ (reverse) carrying the Kpn1 and Sal1 sites. Underlining indicates cleavage sites of restriction enzymes. The FLS2 promoter and FLS2 constructs were constructed by PCR-based cloning in vector pCAMBIA2300. Desired pFLS2:FLS2-2×myc-GFP plasmids were introduced into the fls2 mutant plants by Agrobacterium-mediated transformation.

All chemicals were purchased from Sigma-Aldrich, if not otherwise indicated, and used at the following concentrations: TyrA23 (50 mM in DMSO, working solution 50 µM), TyrA51 (50 mM in DMSO, working solution 50 µM), methyl-β-cyclodextrin (MβCD) (200 mM in water, working solution 10 mM), brefeldin A (BFA) (50 mM in DMSO, working solution 50 µM) and FM4-64 (5 mM in DMSO, working solution 5 µM). The flagellin peptides flg22 and flg22Δ2 were synthesized by GL Biochem (Shanghai) and were used in a concentration of 10 μM (10 mM in double distilled H₂O). The TyrA23, TyrA51, MβCD and BFA treatments were performed as described previously (Wang et al., 2015) and stock solutions were diluted with 1/2 MS growth medium.

Arabidopsis seedlings were imaged with an Olympus FluoView 1000 inverted confocal microscope fitted with a 60× water-immersion objective (numerical aperture 1). GFP/FM4-64 and RFP/mCherry were excited with 488 and 561 nm wavelengths, respectively, and the fluorescence emission spectra were detected at 500-530 nm for GFP, 600-700 nm for FM4-64 and 580-620 nm for RFP/mCherry. The original images were obtained from the same confocal microscope with the same laser and camera settings and then images were processed using the Olympus, Adobe Photoshop CS5 and ImageJ software packages.

Total internal reflection fluorescence microscopy (TIRFM) was performed on an inverted microscope with a total internal reflective fluorescence illuminator and a 100× oil-immersion objective (Olympus; numerical aperture=1.45). For the detection of FLS2-GFP particles at the plasma membrane, living leaf epidermal cells of 5-day-old seedlings were observed under VA-TIRFM. The images were acquired with 100 ms exposure times and time-lapse series images of single particles of FLS2-GFP were taken at up to 100 images per sequence. Proteins labeled with GFP or mCherry were excited with 473-nm or 561-nm laser lines, respectively, from a diode laser (Changchun New Industries Optoelectronics Tech. Co., Changchun, China) and their emission fluorescence was obtained with a filter (BA510IF for EGFP; HQ605/52 for mCherry). The fluorescent signals were acquired by a digital EMCCD camera (ANDOR iXon DV8897D-CS0-VP, Andor Technology, Belfast, UK) after filtering with two band-pass filters (525/545 and 609/654 nm). Images were acquired with 100 ms exposure times and stored directly on a computer. The methods used to analyze single-particle tracking were described previously (Wang et al., 2015).

A program named ‘progressive idealization and filtering’ (PIF) was used for the photobleaching step analysis. After background subtraction by the rolling ball method in ImageJ software, images were input into the PIF program and subjected to the steps described previously (McGuire et al., 2012).

FCS was carried out on a Leica TCS SP5 FCS microscope equipped with a 488 nm argon laser, an in-house coupled correlator and an Avalanche photodiode. After acquiring images of cells in transmitted light mode, FCS was performed in the point-scanning mode. The diffusion of FLS2-GFP molecules into and out of the focal volume changed the local concentration of the fluorophore number, which led to spontaneous fluctuation of the fluorescence intensity. Finally, the FLS2-GFP density of individual cell membranes was calculated according to the protocol described previously (Li et al., 2011). Up to six random points were selected in one cell, five cells were chosen from a leaf and at least three representative leaves were measured for each study.

The images were obtained with a FluoView 1000 laser-scanning microscope (Olympus). The relative surface of the internalized proteins or relative fluorescence intensity at the plasma membrane was calculated with ImageJ software (National Institutes of Health, version 1.48) as previously described (Du et al., 2013).

Total proteins were extracted from 10-day-old seedlings of transgenic FLS2-GFP lines in the fls2 or smt1 mutant backgrounds under different conditions. The GFP antibody (Sigma-Aldrich) was raised against GFP peptides of FLS2-GFP protein. FLS2-GFP protein was denatured in buffer E [100 ml buffer E includes 1.5125 g Tris-HCl (pH 8.8), 11.1 ml glycerine, 1.2409 g Na₂S₂O₅, 5 mM DTT and 1 g SDS]. Proteins were separated with an 8% SDS-polyacrylamide gel and transferred to a nitrocellulose membrane. A 1:4000 dilution was used for the anti-GFP antibody.

The net flux of Ca²⁺ was measured using the non-invasive micro-test technique (NMT) method, as previously described (Yan et al., 2015). This technique is used to study the flux of specific molecules from single cells and/or tissues, and is also known as the self-referencing microelectrode technique. The Ca²⁺ electrode was fabricated as described by Yan (Yan et al., 2015). Pre-pulled and silanized glass micropipettes (2-4 μm aperture) were first filled with a backfilling solution (Ca²⁺: 100 mM CaCl₂) to a length of about 1 cm from the tip. Then, the micropipettes were front-filled with selective liquid ion-exchange (LIX) cocktails to a column length of approximately 25 μm. The LIX cocktails were obtained from Sigma-Aldrich (Sun et al., 2009). An Ag/AgClelectrode holder was inserted from the back of the electrode until the tip contacted the surface of the electrolyte solution. DRIREF-2 (World Precision Instruments) was used as a reference electrode. The electrodes were calibrated prior to the experiment in known calibration solutions (0.05, 0.1, and 0.5 mM Ca²⁺). Only electrodes with Nernstian slopes between 53 and 62 mV/[log-C(±z)] were used in our study (where z is the valence of the ion and C is the concentration of the ion).

Leaves from 7-day-old seedlings were fixed on the bottom of a 35 mm dish incubated in the test buffer. The Ca²⁺ flux of the leaves was then measured using NMT. Each plant was measured once. The leaves were then treated with 0.1 μM flg22 and the Ca²⁺ flux was measured again. The final flux values were reported as the mean of 7 to 11 individual leaves. The Ca²⁺ fluxes were calculated as described by Yan et al. (2015).

Total RNA was extracted from 10-day-old seedlings with a TiangenNApre Plant Kit (Tiangen) according to the manufacturer's instructions. Reverse transcription was performed using FastQuant RT Kit (Tiangen). qPCR was performed with the TiangenSurperRealPreMix Plus (SYBR Green). The flg22-induced early-response genes and primers were listed as follows: WRKY40 (At1g80840), 5′-GATCCACCGACAAGTGCTTT-3′ and 5′-AGGGCTGATTTGATCCCTCT-3′; CYP81f2 (At5g57220), 5′-CTCATGCTCAGTATGATGC-3′ and 5′-CTCCAATCTTCTCGTCTATC-3′.

Detection of callose deposition was performed as previously described (Cao et al., 2013). In brief, 7-day-old seedlings were infiltrated with 2 μM flg22 in six-well plates for 24 h. Control treatments were infiltrated with H₂O. Excised leaves were immediately cleared in 95% ethanol:H₂O (3:1) for at least 6 h. Subsequently, samples were rinsed with 50% ethanol and H₂O. Seedlings were stained with Aniline Blue in 0.15 M phosphate buffer (pH 9.5) and callose deposits were visualized using ultraviolet epifluorescence microscopy as described previously (Cao et al., 2013).

ROS burst experiments were performed by a luminol-based assay as described previously (Sun et al., 2012). Leaves of 4-week-old transgenic Arabidopsis plants were sliced into 1 mm strips and incubated in 100 μl of H₂O in a 96-well plate overnight. The H₂O was then replaced by 100 μl of reaction solution containing 50 μM of luminol and 10 μg/ml of horseradish peroxidase (Sigma-Aldrich) supplemented with 100 nM of flg22. Luminescence was acquired over time using a Photek camera and software. The reported data are pooled measurements from six independent leaves per condition originating from a minimum of three plants.

Protoplasts were isolated and transfected as previously described (Niu and Sheen, 2012). After incubation overnight, transformed protoplasts were treated with water or 1 mM flg22 for 10 min. Total protein was extracted with protein extraction buffer [50 mM HEPES (pH 7.5) 150 mM KCl, 1 mM EDTA, 0.2% Triton X-100, 1 mM DTT, proteinase inhibitor cocktail (Roche)]. For anti-FLAG IP, total protein was incubated with 50 µl agarose-conjugated anti-FLAG antibody (Sigma-Aldrich) for 4 h. After washing six times with extraction buffer. The bound protein was eluted by incubating with 3× FLAG peptide (Sigma-Aldrich) for 1 h. The immunoprecipitates were separated by SDS-PAGE and detected with immunoblotting.

The significance of arithmetic mean values for all data sets was assessed using Student's t-test. Error bars were calculated with the s.d. function in Microsoft Excel. The differences at P<0.05 were considered statistically significant. According to Student's t-test, characters in the figure represent statistically significant differences compared with control (*P<0.05, **P<0.01 and ***P<0.001).

Sequence data used in this article can be found in the Arabidopsis Information Resource (TAIR) database under the following accession numbers: FLS2 (AT5G46330), Flot 1 (AT5G25250) and smt1 (At5g13710).

We sincerely thank Dingzhong Tang (University of Chinese Academy of Sciences) for providing the fls2 mutant line and Timothy Nelson (Yale University) for providing the smt1 mutant line. We also thank Qihua He (Peking University) for technical assistance with FCS.