Mechanosensory trichome cells evoke a mechanical stimuli–induced immune response in Arabidopsis thaliana

Perception of pathogen-derived ligands by corresponding host receptors is a pivotal strategy in eukaryotic innate immunity. In plants, this is complemented by circadian anticipation of infection timing, promoting basal resistance even in the absence of pathogen threat. Here, we report that trichomes, hair-like structures on the epidermis, directly sense external mechanical forces, including raindrops, to anticipate pathogen infections in Arabidopsis thaliana. Exposure of leaf surfaces to mechanical stimuli initiates the concentric propagation of intercellular calcium waves away from trichomes to induce defence-related genes. Propagating calcium waves enable effective immunity against pathogenic microbes through the CALMODULIN-BINDING TRANSCRIPTION ACTIVATOR 3 (CAMTA3) and mitogen-activated protein kinases. We propose an early layer of plant immunity in which trichomes function as mechanosensory cells that detect potential risks.

I nnate immunity is an evolutionarily conserved front line of defense across the plant and animal kingdoms. In plants, pattern-recognition receptors (PRRs), such as leucine-rich repeat receptor-like kinases (LRR-RLKs) and LRR receptor proteins (LRR-RPs), specifically recognize microbe-associated molecular patterns (MAMPs) as non-self molecules, leading to the activation of pattern-triggered immunity (PTI) to limit pathogen proliferation 1,2 . While adapted pathogens have evolved virulence effectors that can circumvent PTI, plants also deploy disease resistance (R) genes, primarily encoding nucleotide-binding LRR proteins, which mount effector-triggered immunity (ETI) [3][4][5] . ETI often culminates in a hypersensitive response as well as acute and localized cell death at the site of infection accompanied by profound transcriptional changes of defense-related genes to retard pathogen growth 4,5 . These ligand-receptor systems are largely dependent on a transient increase in intracellular calcium concentration ([Ca 2+ ] i ), followed by the initiation of phosphorylation-dependent signaling cascades, including mitogen-activated protein kinases (MAPKs) and calcium-dependent protein kinases, that orchestrate a complex transcriptional network and the activity of immune mediators 6,7 .
In addition to PTI and ETI, plant immunity can be induced periodically in the absence of pathogen threat, a process controlled by the circadian clock and driven by daily oscillations in humidity as well as light-dark cycles [8][9][10] . Such responses enable plants to prepare for the potential increased risk of infection at the time when microbes are anticipated to be most infectious. The anticipation of potentially pathogenic microorganisms through sensing climatological changes and their specific detection thus constitute two distinct layers of the plant immune system.
Among the climatological factors that affect the outcome of plant-microbe interactions, rain is a major cause of devastating plant diseases, as fungal spores and bacteria are spread through rain-dispersed aerosols or ballistic particles splashed from neighboring infected plants. Natural raindrops contain bacteria at a concentration of 1.06 × 10 4 (/cm 3 ) 11 , including plant pathogens such as Pseudomonas syringae 12 , Xanthomonas campestris, and Pantoea ananatis 13 . Likewise, raindrops contain fungi such as Alternaria sp., Fusarium sp., Cladosporium sp., Phoma sp., Rhizopus sp., and Botrytis cinerea 14 . In addition, raindrops negatively regulate stomatal closure, which facilitates pathogen entry into leaf tissues 13,15,16 . High humidity, which is usually associated with rain, enhances the effects of bacterial pathogen effectors, such as HopM1, and establishes an aqueous apoplast for aggressive host colonization 17 . These findings suggest that it would be beneficial for plants to recognize rain as an early risk factor for infectious diseases.
How do plants respond to rain? Rain-simulating water spray induces the expression of mechanosensitive TOUCH (TCH) genes as bending stimulation, which induces plant growth retardation 18 . Mechanostimulation affects a variety of plant physiological processes mediated by phytohormones such as auxin, ethylene, and gibberellin [19][20][21][22] . Arabidopsis thaliana seedlings exposed to rainsimulating water spray accumulate the immune phytohormone jasmonic acid (JA) to promote the expression of JA-responsive genes 23 . Thus, rain modulates both mechanotransduction and hormone-signaling pathways that could affect the growth and development of plants as well as environmental responses. However, the regulatory mechanisms underpinning the rain-activated signaling pathway have not been fully elucidated.
Here, we report an early layer of the plant immune system evoked by sensing mechanostimulation on the leaf surface: trichomes, hairlike cells, function as mechanosensory cells that mount an effective immune response against both biotrophic and necrotrophic pathogens. When trichomes are mechanically stimulated, intercellular calcium waves are concentrically propagated away from them, MAPKs are activated, and the calcium-and calmodulin-binding transcription activator (CAMTA)-regulated immune response is initiated. We propose that plants directly recognize rain as a risk factor and evoke a rapid immune response that substantially contributes to early detection of, and protection from, potential pathogens.

Results and discussion
Rain and mechanical stimuli induce mechanosensitive genes involved in plant immunity. To investigate the effect of rain on transcriptional changes in Arabidopsis leaves, we performed transcriptome deep sequencing (RNA-seq) of wild-type Columbia (Col-0) Arabidopsis treated with artificial raindrops (Supplementary Fig. 1a; Methods). After applying only 10 falling droplets, we detected the marked induction of 1050 genes 15 min after treatment [3 biological replicates, log 2 fold changes (log 2 FC) ≥ 1, likelihood ratio test; P < 0.05] (Supplementary Data 1). Gene Ontology (GO) analysis of these genes revealed a striking enrichment in categories associated with plant immunity, as evidenced by the expression of major immune regulators, including WRKY DNA-binding protein (WRKY) genes, calmodulin-binding protein 60-like g (CBP60g), MYB domain protein (MYB) genes, ethylene response factor (ERF) genes, and MAP kinase (MPK) genes 24,25 (Fig. 1a and Supplementary Data 1, 2). The touchinduced genes TCH2 and TCH4 were also highly upregulated in response to one falling raindrop (falling) compared to a water droplet placed directly on the leaf surface (static) ( Supplementary  Fig. 1b). These results suggested that mechanosensation is involved in altering transcriptional activity.
To validate this hypothesis, we mechanically stimulated rosette leaves by gently brushing them 1-10 times along the main veins with a small paintbrush (Supplementary Fig. 1c; Methods) and analyzed the expression profile of the immune regulator WRKY33, which was responsive to raindrops. WRKY33 expression was maximally induced 15 min after brushing the leaves one to four times ( Supplementary Fig. 1d). Next, we compared gene expression patterns between leaves that were brushed once and those that received 10 falling raindrops. Both raindrops and brushing strongly upregulated TCH2, TCH4, WRKY33, WRKY40, WRKY53, CBP60g, MYB51, ERF1, and jasmonate-ZIM-domain protein 1 (JAZ1) expression 18,24,25 (Fig. 1b, c), suggesting that raindrops are likely recognized as a mechanical stimulus. This idea was supported by the result that natural rainfall-induced the expression of TCH4 and WRKY33 in Col-0 leaves (Supplementary Fig. 1e).
To comprehensively identify mechanosensitive genes, we performed RNA-seq analysis of leaves brushed once. We identified 1241 genes that were significantly induced 15 min after this treatment relative to control plants (3 biological replicates, log 2 FC ≥ 1, likelihood ratio test; P < 0.05) (Supplementary Data 3). These mechanical stimuli (MS)-induced genes were primarily categorized as plant immune responses, such as response to chitin, defense response, and immune system response ( Fig. 1d and Supplementary Data 4). We found that 87.3% of raindropinduced genes and 73.9% of MS-induced genes overlapped (Fig. 1e): this set of 917 genes expressed upon both treatments were enriched for GO categories associated with stress responses ( Supplementary Fig. 2). Furthermore, the expression levels of these 917 genes, including major immune regulators, were positively correlated between the two treatments (Pearson correlation coefficient r = 0.917) (Fig. 1f). In addition, we observed clear correlations between our profiles and previously published datasets obtained by water spray 23 , bending 26 , brushing 27 , and cotton swabbing treatments 28 (Supplementary Fig. 3 and Supplementary Data 11). Strong correlations observed in Fig. 1f are consistent with the fact that 10 falling droplets ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-28813-8 applied the same level of the force intensity as brushing the leaf surface once (Supplementary Fig. 4). Taken together, these transcriptome analyses indicated that falling raindrops stimulate the expression of mechanosensitive genes involved in environmental stress responses, including plant immunity.
Rain and MS rapidly activate plant immune responses. To further characterize raindrop-induced genes, we conducted a comparative analysis with published transcriptome datasets. Many raindrop-and MS-induced genes were also expressed during major plant immune responses, such as those triggered by the immune phytohormones salicylic acid (SA), which is effective against biotrophic pathogens (21%; 193/917 genes), and JA, which mounts immune responses to necrotrophic pathogens (11.8%; 108/917 genes); the bacterial-derived peptide flg22, which activates PTI (37%; 339/917 genes); and the bacterial pathogen P. syringae pathovar maculicola ES4326 (Psm ES4326) (25.8%; 237/ 917 genes) 1,2,29-33 (Fig. 2a, b). In total, 58.6% (537/917 genes) of raindrop-and MS-induced genes overlapped with those induced in response to different immune elicitors, suggesting that raindrops activate mechanosensitive immune responses.    genes overlap with SA-responsive genes (Fig. 2a), SA levels were not significantly increased in response to raindrops and MS (Fig. 2c). Mechanosensitive genes such as TCH2, TCH4, WRKY33, WRKY40, CBP60g, MYB51, ERF1, and JAZ1, whose expression was induced 15 min after treatment with MS and raindrops, are presumably regulated independently of phytohormonal responses, as the expression of these genes was not compromised in mutants of the JA-responsive transcription activators, MYC2, MYC3, and MYC4, and the JA-insensitive jai3-1 (JASMONATE-INSENSITIVE 3) (Supplementary Fig. 6). A previous report demonstrated that GA accumulation is reduced by "bending" leaves twice per day for 2 weeks 22 . Here, significant changes in GA levels were not detected upon transient application of raindrops or MS ( Supplementary  Fig. 5). Therefore, mechanotransduction may branch into phytohormone-dependent and -independent pathways, which are differentially activated by MS depending on their type, intensity, and duration and on which organs perceive the stimulation.
Activation of MAPKs is one of the earliest cellular events and a hallmark of plant immune responses. In particular, PRRs promptly activate a phosphorylation cascade involving MPK3 and MPK6 in response to MAMPs, whereby the downstream immune components of PTI are phosphorylated to promote transcriptional reprogramming 1,6,7,34 . Because 37% of raindropand MS-induced genes were also upregulated by flg22 treatment (Fig. 2a), we examined whether a MAPK cascade is activated in responses to raindrops and MS by immunoblot analysis with the anti-p44/42 antibody, which detects phosphorylated MPK3/ MPK6 35,36 . A previous study found that MPK4/MPK6 phosphorylation was induced by bending 28,37 . Upon treatment of rosette leaves with four falling raindrops or MS (four brushing), phosphorylation of MPK3/MPK6 was induced within 3 min and remained high for 10 min after each treatment (Fig. 2d, e), indicating that MPK3/MPK6 activation precedes the expression of mechanosensitive genes detected 10 min after MS application ( Supplementary Fig. 1d). The kinetics of MS-activated MPK3/ MPK6 were reminiscent of those observed upon activation of the PRR protein FLAGELLIN-SENSITIVE 2 (FLS2) and its coreceptor BRI1-associated receptor kinase 1 (BAK1), which are responsible for recognition of the bacterial flg22 epitope 1,34 . Wild-type, fls2, and bak1 mutant plants displayed comparable levels of phosphorylated MPK3/MPK6 in response to MS (Fig. 2f), however, suggesting that FLS2 and BAK1 are not positively involved in raindrop-elicited mechanotransduction.
We then performed a comparative analysis of raindrop-and MS-induced genes against published transcriptome datasets Rain and MS confer resistance to both biotrophic and necrotrophic pathogens. We then investigated whether raindrops and MS confer resistance to pathogenic microbes. Raindrops containing the spores of the necrotrophic pathogen Alternaria brassicicola Ryo-1 were placed on fully expanded leaves after pretreatment with raindrops or MS for 3 h at an interval of 15 min. Both stimuli significantly suppressed lesion development compared to control plants without pretreatment (Fig. 2g, h). Pretreatment of leaves with MS for 3 h also efficiently protected plants from infection with the biotrophic pathogen Psm ES4326 (Fig. 2i). These results confirmed that mechanostimulation induces a PTI-like response to confer a broad spectrum of resistance to both biotrophic and necrotrophic pathogens, as MS activates immune MAPKs and upregulates a large subset of flg22induced genes. In support of this argument, exposure to the fungal cell wall, chitin, also upregulated 42.1% (386/917 genes) of raindrop-induced genes ( Supplementary Fig. 7c).
Mechanosensitive genes are regulated by calmodulin-binding transcription activator 3. To dissect rain-induced mechanotransduction, we searched for a conserved cis-regulatory element in the promoter sequences of mechanosensitive genes. From an unbiased promoter analysis of the top 300 genes among 917 differentially expressed genes, we obtained the highest enrichment for the CGCG box (CGCGT or CGTGT), which is recognized by CAMTAs that are conserved from plants to mammals [38][39][40][41][42] (Fig. 3a). A similar motif analysis that detects the CAMTA-binding sites among the brushing-induced gene promoters was reported 27 . The Arabidopsis transcription factor CAMTA3 [also named signal responsive 1 (SR1)] is a negative regulator of plant immunity; camta3 null mutants exhibit constitutive expression of defense-related genes and enhanced resistance to virulent P. syringae infection 43,44 . CAMTA transcription Venn diagram (a) and the upset plot (b) between 917 raindropand MS-induced genes and transcriptome datasets obtained from salicylic acid (SA), jasmonic acid (JA), and flg22 (PAMP) treatment and Pseudomonas syringae pv. maculicola ES4326 infection (P < 0.05). Overlap with raindrop-and MS-induced genes: SA, 21%, 193/917 genes; JA, 11.8%, 108/917 genes; flg22, 37%, 339/917 genes; Psm ES4326, 25.8%, 237/917 genes; any of the four factors, 58.6%, 537/917 genes. c Accumulation of plant hormones (ng/g fresh weight) SA, JA, JA-isoleucine (JA-Ile), abscisic acid (ABA), and indole-3-acetic acid (IAA) 5 min after treatment with 10 falling droplets (raindrop), 1 brushing (MS), or cutting (wounding). Data are presented as mean ± SD. n = 6 plants examined over three independent experiments. Each dot indicates a biological replicate. Different letters above bars indicate significant differences (one-sided Tukey's multiple comparison test; P < 0.05). d, e Raindrop (4 droplets)-(d) and MS (4 brushings)-induced (e). MAPK activation in Col-0. Total proteins were extracted from 4-week-old plants treated with raindrops and detected by immunoblot analysis with anti-phospho-p44/42 MAPK antibodies. Relative phosphorylation levels are shown below each blot. Similar results were obtained in four independent experiments. f MS-induced MAPK activation in Col-0, fls2, and bak1-3. Total proteins were extracted from 4week-old plants after 5 min of MS treatment (1 brushing) and detected by immunoblot analysis with anti-phospho-p44/42 MAPK antibodies. Relative phosphorylation levels are shown below each blot. Similar results were obtained in three independent experiments. g, h Disease progression of Alternaria brassicicola in Col-0 leaves 3 days after inoculation with (+) or without (−) raindrop (10 falling droplets) pretreatment (g) or with (+) or without (−) MS (4 brushings) pretreatment (h). Error bars represent SE. Asterisks indicate significant difference (two-sided Tukey's t test; ****P < 0.0001). i Growth of Psm ES4326 in Col-0 leaves 2 days after inoculation with (+) or without (−) MS (4 brushings) pretreatment. Error bars represent SE. Asterisks indicate significant difference (two-sided Tukey's t test; ****P < 0.0001). Cfu colony-forming units. factors possess a CaM-binding domain and an IQ domain to which CaM binds in a calcium-dependent manner to negate their function ( Supplementary Fig. 8a). CAMTA3 A855V transgenic plants, which possess a mutation in the IQ domain, suppress the constitutive expression of defense-related genes seen in the camta2 camta3 double mutant and are no longer regulated by calcium-mediated responses 45,46 . In agreement with our promoter analysis, 28.7% of constitutively upregulated genes (309/ 1075 genes) in the camta1 camta2 camta3 triple mutant overlapped with raindrop-and MS-induced genes detected in wildtype plants 47 ( Supplementary Fig. 8b and Supplementary Data 7). Upon application of raindrops and MS, WRKY33 and CBP60g transcript levels were significantly reduced in plants expressing the CAMTA3 A855V variant compared to a CAMTA3-GFP transgenic line expressing a transgene that complemented the phenotype of the camta2 camta3 mutant (Fig. 3b, c), suggesting that CAMTA3 is involved in mechanotransduction.
To confirm whether CAMTA3 directly targets mechanosensitive genes, we investigated the genome-wide distribution of CAMTA3-binding sites by chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) using CAMTA3 A855V -GFP plants, as the mutant protein stably represses the transcription of CAMTA3-regulated genes. With the aid of model-based analysis of ChIP-seq (MACS2) software, we identified 2641 and 2728 CAMTA3-binding genes, respectively, in two replicates (binomial distribution; P < 0.05); about 40% of these peaks are located in the promoter regions and another 30% in gene bodies (Supplementary Fig. 8c and Supplementary Data 8). The overlap between the two replicates highlighted 2011 CAMTA3-targeted genes that included 272 raindrop-and 297 MS-induced genes such as TCH2, TCH4, and CBP60g (Fig. 3d), consistent with our hypothesis that CAMTA3 regulates the transcription of mechanosensitive genes.
To validate the results from the promoter analysis of mechanosensitive genes, we next investigated specific DNA sequences to which CAMTA3 selectively binds by analyzing CAMTA3-binding peaks by Multiple EM for Motif Elicitation (MEME)-ChIP (Methods). We again identified the CGCG box (CGCGT or CGTGT) as the motif with the highest enrichment score (3.4 × 10 −44 ) (Fig. 3e). Subsequent visualization of ChIPseq profiles via the Integrative Genomics Viewer (IGV) 48 demonstrated that CAMTA3 is primarily enriched at the CGCG boxes of mechanosensitive genes, including TCH2, TCH4, CAM2, CBP60g, calmodulin-like 23 (CML23), and WRKY40 (Fig. 3f). GO analysis on 314 CAMTA-targeted genes (Fig. 3d, shown in red) to define the biological functions of these genes showed significant enrichment in categories related to immune and environmental responses ( Supplementary Fig. 8d, e). We thus investigated whether CAMTA3 is required for the immune responses. camta2 camta3 CAMTA3-GFP transgenic plants effectively mounted an enhanced disease resistance against P. syringae in response to MS, while camta2 camta3 CAMTA3 A855V -GFP plants were significantly compromised in resistance (Fig. 3g). These results demonstrate that CAMTA3 negatively regulates plant immune responses by binding to the CGCG box in raindrop-and MS-induced gene promoters and represses the expression of these genes.
Since mechanostimulation rapidly activates MPK3/MPK6 (Fig. 2d, e), we investigated whether CAMTA3 mediates the activation of these MPKs. Using camta2 camta3 CAMTA3-GFP and camta2 camta3 CAMTA3 A855V -GFP, we detected the phosphorylation of MPK3/MPK6 independently of CAMTA3 activity (Supplementary Fig. 8f). In addition, the calcium ionophore A23187 clearly induced the phosphorylation of MPK3 and MPK6 ( Supplementary  Fig. 8g). These results suggested that the mechanotransduction initiated by raindrops and MS may cause a Ca 2+ influx that negates the repressive effect of CAMTA3 and independently activates the MAPK cascade, as previously proposed 34 .
MS initiates intercellular calcium waves concentrically away from trichomes. To visualize how mechanostimulation induces the expression of immune genes in planta, we generated Arabidopsis transgenic lines with the promoter sequences of WRKY33 and CBP60g driving the expression of nucleus-targeted enhanced yellow fluorescent protein (YFP-NLS) (WRKY33pro:EYFP-NLS and CBP60gpro:EYFP-NLS). WRKY33 expression is regulated by both MPK3/MPK6 and CAMTA3, while CBP60g is not mediated by MPK3/MPK6 (Supplementary Data 5). When half leaves were gently brushed ( Supplementary Fig. 1c), we detected YFP fluorescence in the WRKY33pro:EYFP-NLS and CBP60gpro:EYFP-NLS transgenic lines as localized, clustered groups of cells only in the brushed half ( Fig. 4a and Supplementary Fig. 9a). Closer inspection of the stimulated regions revealed that both genes were induced in cells surrounding trichomes, hair-like structures projecting outward from the epidermal surface (Fig. 4b, c and Supplementary Fig. 9b, c).
Trichomes function as chemical and physical barriers against insect feeding and are likely involved in drought tolerance and protection against ultraviolet irradiation 49,50 . Mechanostimulation of a single trichome induces Ca 2+ oscillations within the proximal skirt cells that surround the base of trichomes 51 , suggesting that the mechanical force could be focused on only skirt cells (Supplementary Fig. 10). However, since mechanostimulation by raindrops and MS confers resistance to pathogens in whole leaves, we hypothesized that trichomes activate a Ca 2+ signal in a large area of leaves, as shown in Fig. 4a.
To visualize changes in cytosolic Ca 2+ concentrations ([Ca 2+ ] cyt ) induced by MS on the leaf surface, we used transgenic Arabidopsis expressing the GFP-based [Ca 2+ ] cyt indicator GCaMP3 52,53 . Leaf brushing induced a marked increase of [Ca 2+ ] cyt in the surrounding leaf area of trichomes 1 min after stimulation ( Fig. 4d and Supplementary Movie 1). Flicking a single trichome with a silver chloride wire triggered an intercellular calcium wave that propagated concentrically away from the trichome and surrounding skirt cells at a speed of 1.0 μm/s (Fig. 4e, f and Supplementary Movie 2). This pattern showed striking consistency with the area of induction observed with the WRKY33pro:EYFP-NLS and CBP60gpro:EYFP-NLS reporters (Fig. 4a, b and Supplementary Fig. 9a, b). The base of trichomes exhibited a rapid and transient increase in [Ca 2+ ] cyt before the concentric propagation of calcium waves was initiated ( Fig. 4g and Supplementary Movie 3).
Trichomes are mechanosensory cells activating plant immunity. To investigate the possible involvement of trichomes in mechanosensation in Arabidopsis leaves and activation of the immune response, we observed calcium waves using the knockout mutant of glabrous 1 (GL1) 54 , which lacks trichomes. The gl1 mutant exhibits effective basal resistance comparable to that of wild-type Col-0 plants 54,55 , and its local resistance to Psm ES4326 and A. brassicicola Ryo-1 is similar to that of Col-0 plants ( Supplementary Fig. 11). However, the trichome-defective gl1, gl3 (glabrous 3), and ttg1 (transparent testa glabra 1) mutants lost cuticle and systemic acquired resistance (SAR) 55 , a secondary immune response in uninfected distal tissues after primary infection in a local site, while basal levels of immunity were not dramatically affected 55 (Fig. 5 and Supplementary Fig. 11). The lack of SAR in these trichome mutants is associated with impairment of cuticle formation but not of trichomes. cer1-1 (eceriferum 1) and cer3-1 (WAX 2/eceriferum 3) mutants, which have defects in synthesizing cuticular wax but form normal trichomes, fail to induce SAR 56 . Therefore, the immunodeficient phenotype of gl and ttg mutants is due to disordered cuticle formation. To investigate whether trichome-mediated immunity is compromised in cer1 and cer3 mutants as in gl1, we performed the pathogen test using A. brassicicola Ryo-1 after mechanical stimulation of these mutants ( Supplementary Fig. 12). Pretreatment with brushing significantly suppressed lesion development. These data strongly indicate that the lack of trichome, but not cuticle, in gl mutants is responsible for failed MS-induced immunity. The mechanostimulation-induced propagation of concentric calcium waves was compromised in the gl1 mutant (Fig. 5a, b and Supplementary Fig. 13), confirming that trichomes are true MS sensors and initiate calcium waves (Supplementary Movies 4 and 5). Furthermore, approximately 70.5% of mechanosensitive genes were expressed in a trichome-dependent manner (3 biological replicates, log 2 FC ≥ 1, likelihood ratio test; P < 0.05) (Fig. 5c, Supplementary Fig. 14a and Supplementary Data 9), and transcript levels of 18 representative mechanosensitive immune genes were markedly lower at all time points in the gl1 mutant than they were in the wild type in RNA-seq d c e Col-0  (Fig. 5d), suggesting that trichomes serve as a sensor of this type of MS on the leaf surface. Similarly, compared to wild-type plants, the transcription of MSand raindrop-induced WRKY33, WRKY40, and CBP60g, as well as the activation of MPK3/MPK6 by MS, was also significantly reduced in the gl1 mutant (Fig. 5e, f and Supplementary Fig. 14b). Moreover, MS-induced resistance against Psm ES4326 infection was abrogated in the gl1 mutant (Fig. 5g). As with Psm ES4326, the application of MS to wild-type plants prior to inoculation with A. brassicicola significantly limited lesion development, whereas the gl1 mutant did not show enhanced disease resistance in response to MS (Fig. 5h). Mechanically stimulating leaves by bending them back and forth 10 times induces JA-dependent resistance to B. cinerea in both Col-0 and gl1 mutant 57 , suggesting that bending directly stimulates cells inside the leaf.
Our work highlights an early layer of plant immunity that is triggered by an unexpected function of trichomes as mechanosensory cells. When trichomes are mechanically stimulated, intercellular calcium waves are concentrically propagated away from the trichomes, followed by the activation of CAMTA3dependent immune responses (Fig. 6). Rapid phosphorylation of MAPKs also is a prerequisite for mechanosensitive gene expression 28 , as MPK3/MPK6 mediate the phosphorylation of mechanosensitive WRKY33 for its activation 58,59 . This notion is Fig. 5 Trichomes are mechanosensory cells. a Ca 2+ imaging using 35Spro:GCaMP3 (Col-0) and 35Spro:GCaMP3 (gl1). Leaf surfaces were exposed to MS by brushing. MS-induced calcium waves were compromised in the gl1 mutant. See also Supplementary Movies 4 and 5. Scale bars, 0.5 mm. b [Ca 2+ ] cyt signature of (a). c Venn diagram of transcriptome datasets for MS-induced genes in Col-0 and gl1 (likelihood ratio test; P < 0.05). NS, not significant. Lower, fold change (FC) (gl1)/FC (Col-0) < 0.5. High, MS (gl1)/Mock (gl1), log 2 FC ≥ 1 in gl1 (likelihood ratio test; P < 0.05). d Heatmap of differentially expressed defense-related genes obtained from transcriptome datasets from Col-0 and gl1 plants treated with MS (4 brushings). e Transcript levels of WRKY33, WRKY40, and CBP60g in 4-week-old Col-0 and gl1 plants 15 min after treatment with brushing 4 times, determined using RT-qPCR and normalized to UBQ5. Data are presented as mean ± SD. Asterisks indicate significant difference (one-sided Tukey's multiple comparison test; ***P < 0.001, ****P < 0.0001). n = 6 plants examined over three independent experiments. Each dot indicates a technical replicate. f MS-induced MAPK activation in Col-0 and gl1. Total proteins were extracted from 4-week-old leaves 5 min after MS (4 brushings) treatment and detected by immunoblot analysis with anti-phospho-p44/42 MAPK antibodies. Relative phosphorylation levels are shown below each blot. Similar results were obtained in three independent experiments. g Growth of Psm ES4326 in Col-0 and gl1 leaves 2 days after inoculation with (+) or without (−) MS (4 brushings) pretreatment. Error bars represent SE. Asterisks indicate a significant difference (one-sided Tukey's t test and two-way ANOVA; ***P < 0.001). Cfu colony-forming units, NS not significant. n = 8 samples examined over three independent experiments. Each dot indicates a biological replicate. h Disease progression of Alternaria brassicicola in Col-0 and gl1 leaves 3 days after inoculation with (+) or without (−) MS (4 brushings) pretreatment. Error bars represent SE. Asterisks indicate a significant difference (one-sided Tukey's t test and two-way ANOVA; ****P < 0.0001). NS not significant. Col-0: n = 15, gl1: n = 14 samples examined over three independent experiments. Each dot indicates a biological replicate.  supported by the finding that the expression of 252 of 917 raindrop-and MS-induced genes are mediated by MPK3/MPK6 ( Supplementary Fig. 7a), and their promoter sequences possess the W-box (TTGACC) for WRKYs as the most enriched cisregulatory elements (Supplementary Fig. 15). The molecular mechanism by which calcium mediates the activation of MPK3/ MPK6 has yet to be elucidated.
Mechanostimulation by repeatedly bending leaves confers resistance to the necrotrophic pathogen B. cinerea via JA accumulation 57 . In addition, a subset of JA-responsive genes upregulated by water spray is mediated by MYC2/MYC3/MYC4 transcription factors 23 . These observations strongly indicate that mechanosensation causes profound JA-dependent changes in gene transcription, promoting plant immune responses to necrotrophic pathogens. The JA-and MYC-dependent pathway does not play a major role in the expression of mechanosensitive TCH genes, however, indicating that mechanotransduction is regulated by other signaling pathways. In addition, we investigated the requirement for MS-induced immunity in the defense mutants fls2, bak1-3, npr1-3 (nonexpresser of PR genes 1), and myc234, and found that FLS2 and NPR1 could be involved in the signaling pathway ( Supplementary Fig. 16). Our work demonstrated that raindrops and MS only partially activate the JA signal but rather strongly induce a PTI-like response via the Ca 2+ -and CAMTA3-dependent pathway, which is highly effective against both necrotrophs and biotrophs (Fig. 2g-i). Because rain disseminates diverse pathogens with different parasitic strategies, including fungi, bacteria, and viruses 60,61 , it is reasonable that plants perceive raindrops as a risk factor and activate broadspectrum resistance.
Plants possess mechanosensory cells with a variety of functions, such as flower antennas of Catasetum species for pollination, tentacles of Drosera rotundifolia for insect trapping, root hairs of Arabidopsis for water tracking, and red cells of Mimosa pudica for evading herbivores 62 . The carnivorous Venus flytrap (Dionaea muscipula) captures insects by sensing mechanostimulation via sensory hairs on leaf lobes 63 . To monitor diverse MS applied to plants, several sensing mechanisms have been proposed that include detection of cell wall components, distortion of the plasma membrane, and displacement of the plasma membrane against the cell wall 64 . In all these systems, a transient increase in [Ca 2+ ] cyt is thought to play a pivotal role in short-and long-term responses. Two successive stimulations of sensory hairs of the flytrap are required to meet the threshold of [Ca 2+ ] cyt for rapid closure of the leaf blade 63,65 . As the trichome on the leaf surface is widely found in many land plants, there may be a common intercellular network of cell-cell communication that initiates calcium waves for activating immune responses.

Plants.
A. thaliana accession Columbia-0 (Col-0) was the background for all plants used in this study. WRKY33pro:EYFP-NLS (Col-0) and CBP60gpro:EYFP-NLS (Col-0) were generated as described in the subsection "Promoter-reporter imaging" 66 . 35Spro:GCaMP3 (Col-0) 53 , camta2 camta3 CAMTA3pro:CAMTA3-GFP, and camta2 camta3 CAMTA3pro:CAMTA3 A855V -GFP were provided from Michael F. Thomashow (Michigan State University) 45,47 . gl1 [Col(gl1)] was obtained from Lehle Seeds (TX, USA) and was previously designated as gl-1 mutant. The Arabidopsis mutants fls2 (SALK_093905) and bak1-3 (SALK_034523) were obtained from the Arabidopsis Biological Resource Center (ABRC). Artificial raindrop treatment. Reverse osmosis water was kept in a 500 mL beaker until the water temperature reached room temperature (22°C). A transfusion set (NIPRO Infusion Set TI-U250P, Nipro, Osaka, Japan) was installed on a steel stand with the beaker at a height of 1.2 m (H-type Stand I3, As One, Osaka, Japan) and was adjusted to release 13 µL water droplets (Supplementary Fig. 1a). In this setting, the applied mechanical energy to the leaf surface is equivalent to one in which 5.8 µL of raindrops reach a terminal velocity of 6.96 m/s 68 . This size raindrop is frequently observed in nature; thus, the impact of simulated rain is comparable with that of true rain 68 . The adaxial side of leaves from 4-week-old plants was treated with 10 droplets for RNA-seq and 1, 4, or 10 droplets for quantitative RT-PCR (RT-qPCR). The adaxial side of leaves from 4-week-old plants was treated with 1 falling or static droplet (Supplementary Fig. 1b). Sample leaves were collected 15 min after treatment and stored at −80°C until use.
Brush treatment. The adaxial side of leaves from 4-week-old plants was brushed once or 4 times for RNA-seq and RT-qPCR along the main veins at an angle of 30-40°(KOWA nero nylon drawing pen flat 12, Kowa, Aichi, Japan) (Supplementary Fig. 1c). Sample leaves were collected 15, 30, and 60 min after treatment for RNA-seq and 15 min after treatment for RT-qPCR and stored at −80°C until use.
RNA-seq library construction. Total RNA was extracted from 80 to 100 mg frozen samples using Sepasol-RNA I Super G (Nacalai Tesque, Kyoto, Japan) and the TURBO DNase free kit (Thermo Fisher Scientific, IL, USA) according to the manufacturer's protocols. Total RNA was further purified with the RNeasy RNA Isolation Kit (QIAGEN, Hilden, Germany) and assessed for quality and quantity with a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific). We used 1 µg total RNA for mRNA purification with NEBNext Oligo d(T) 25  For the comparative analysis of differentially expressed genes between leaves in the gl1 mutant and Col-0, the amount of cDNA was determined by the QuantiFluor dsDNA System (Promega, WI, USA). cDNA libraries were sequenced as single-end reads for 36 nucleotides on an Illumina NextSeq 500 (Illumina). The reads were mapped to the A. thaliana reference genome (TAIR10) via Bowtie2 70 with the options "--all --best --strata". Pairwise comparisons between samples were performed with the EdgeR package in the R program 69 . Enrichment of GO categories for biological processes was determined using BiNGO (http:// www.psb.ugent.be/cbd/papers/BiNGO/Home.html) (one-sided hypergeometric test; P < 0.05) 71 . Pearson correlation coefficient of the expression levels between two transcriptome profiles was calculated with Excel function (PEARSON function) (r = 0-0.2: no correlation, r = 0.2-0.4: weak correlation, r = 0.4-0.7: slightly correlated, r = 0.7-1.0: correlated).
Quantification of the force density. The abaxial side of leaves from 4-week-old plants was physically attached to the measuring pan of the electronic balance QUINTIX224-1S (Sartorius Lab Instruments GmbH & Co., Göttingen, Germany) with surgical tape (3 M Company, MN, USA). The adaxial side of the leaf was treated with 1 falling droplet or brushed once (shown in the above subsections). The peak weight applied to the leaf surface was obtained as the force. The force per unit area (N/m 2 ) is converted from the peak weight (kg) and the contact area of the brush tip (5.6 × 10 −5 m 2 ) or raindrop (9.73 × 10 −6 m 2 ).
Re-analysis of immune-related transcriptome datasets. We used the following public transcriptome datasets for comparative analysis with the RNA-seq data obtained in this study: 10-day-old Arabidopsis seedlings treated with 1 µM flg22 for  23 , leaves from 4week-old Arabidopsis plants treated with bending back and forth manually for 30 min (accession is now not available) 26 , leaves from 4-week-old Arabidopsis plants treated with brushing for 30 min (NCBI BioProject; PRJNA473032) 27 , 2week-old Arabidopsis seedlings grown with MS medium and treated with cotton swabbing for 30 min (accession is now not available) 28 , 12-day-old Arabidopsis GVG-NtMEK2 DD seedlings treated with 2 μM DEX for 0 and 6 h (NCBI Sequence Read Archive; SRP111959) 36 , and 4-week-old Arabidopsis camta1 camta2 camta3 triple mutant (GSE43818) 47 (Supplementary Data 11). The overlaps between differentially expressed genes in each transcriptome dataset were evaluated as Venn diagrams (http://bioinformatics.psb.ugent.be/webtools/Venn/) and Upset plot (https://asntech.shinyapps.io/intervene/).

RT-qPCR.
Total RNA was extracted from 30-40 mg leaf tissues with Sepasol-RNA I Super G and the TURBO DNase free kit (Thermo Fisher Scientific) according to the manufacturer's protocols, followed by reverse transcription with the Prime-Script RT reagent kit (Takara Bio, Shiga, Japan) using oligo dT primers. RT-qPCR was performed on the first-strand cDNAs diluted 20-fold in water using KAPA SYBR FAST qPCR Master Mix (2×) kit (Roche, Basel, Switzerland) and genespecific primers in a LightCycler 96 (Roche). Primer sequences are listed in Supplementary Data 10.
Quantification of plant hormones. The adaxial side of leaves from 4-week-old plants was treated with 10 raindrops (raindrop), brushed once (MS), or cut (wounding). Sample leaves (0.07-0.1 g) were collected 5 min or 15 min after treatment and stored at −80°C until use. Frozen samples were ground to a fine powder, mixed with 4 mL extraction solution [80% (v/v) acetonitrile, 1% (v/v) acetic acid], and stored for 1 h at 4°C to extract plant hormones. After centrifugation at 3000 g for 10 min, the supernatants were evaporated in a vacuum centrifugal evaporator EC-57CS (Sakuma, Tokyo, Japan) and dissolved in 1% (v/v) acetic acid. Samples were loaded onto a reverse-phase solid-phase extraction cartridge (Oasis HLB 1 cc; Waters Corporation, MA, USA). The cartridge was washed with 1 mL 1% acetic acid, and hormones were eluted with 2 mL extraction solution. The eluent was evaporated to obtain samples in 1 mL 1% acetic acid and subjected to cation exchange chromatography on an Oasis MCX 1 cc extraction cartridge (Waters Corporation). The acidic fraction was eluted with 1 mL extraction solution. The acidic eluent was analyzed for SA, and the remaining fraction was evaporated, dissolved in 1% acetic acid, and loaded onto an Oasis WAX 1-cc extraction cartridge (Waters Corporation). The cartridge was washed with 1% acetic acid and the remaining hormones were eluted with extraction solution. The elute was analyzed for JA, JA-Ile, ABA, IAA, and GA 4 . The contents of these hormones were quantified using liquid chromatography-electrospray tandem mass spectrometry (LC-ESI-MS/MS) (triple quadrupole mass spectrometer with 1260 high-performance LC, G6410B; Agilent Technologies Inc., CA, USA) equipped with a ZORBAX Eclipse XDB-C18 column (Agilent Technologies Inc.) 72 .
ChIP assay. Approximately 0.7 g of 2-week-old camta2 camta3 CAMTA3pro:-CAMTA3 A855V -GFP seedlings was fixed in 25 mL 1% formaldehyde under vacuum for three cycles of 2 min ON/2 min OFF using an aspirator (SIBATA, Tokyo, Japan). Subsequently, 1.5 mL of 2 M glycine was added to quench the cross-linking reaction under vacuum for 2 min. The samples were then washed with 50 mL double-distilled water and stored at −80°C until use. Frozen samples were ground to a fine powder with a mortar and pestle in liquid nitrogen and dissolved in 2.5 mL nuclei extraction buffer (10 mM Tris-HCl pH 8.0, 0.25 M sucrose, 10 mM MgCl 2 , 40 mM β-mercaptoethanol, protease inhibitor cocktail) 33 linking, followed by digestion with 1 μL Proteinase K (20 mg/ml) (Invitrogen, CA, USA) at 37°C for 1 h. ChIP samples were mixed with 500 μL Buffer NTB and purified using the PCR clean-up gel extraction kit following the manufacturer's instructions (MACHEREY-NAGEL, Düren, Germany).
ChIP-seq library construction. ChIP-seq libraries for the input and two biological replicates were constructed from 2 ng purified DNA samples with the NEB Ultra II DNA Library Prep Kit for Illumina (New England Biolabs) according to the manufacturer's instructions. The amount of DNA was determined on an Agilent 4150 TapeStation System (Agilent). All ChIP-seq libraries were sequenced as 81nucleotide single-end reads using an Illumina NextSeq 550 system.
Analysis of ChIP-seq. Reads were mapped to the Arabidopsis thaliana reference genome (TAIR10, http://www.arabidopsis.org/) using Bowtie2 with default parameters 70 . The Sequence Alignment/Map (SAM) file generated by Bowtie2 was converted to a Binary Alignment/Map (BAM) format file by SAMtools 74 . To visualize mapped reads, Tiled Data Files file were generated from each BAM file using the igvtools package in the IGV 48 . ChIP-seq peaks were called by comparing the IP with the Input using MACS2 with the "-p 0.05 -g 1.19e8" option (binomial distribution; P < 0.05) 75 . The peaks were annotated using the nearest gene using the Bioconductor and the ChIPpeakAnno packages in the R program, from which we identified 2011 genes detected in both biological replicates. Enrichment of GO categories of the set of 314 genes overlapping between raindrop-and MS-induced genes for biological processes was determined using BiNGO (http:// www.psb.ugent.be/cbd/papers/BiNGO/Home.html) 71 . Sequences of the peaks were extracted from the Arabidopsis thaliana genome as FASTA files with BEDtools 76 . To identify the candidates of CAMTA3-binding motifs, the FASTA files were subjected to MEME-ChIP with the default parameters (-meme-minw 6-mememaxw 10) 77 , and a density plot of the distribution of the motifs were generated.
Immunoblot analysis for detection of MPK3 and MPK6 phosphorylation. The adaxial side of leaves from 4-week-old plants was brushed four times or treated with four raindrops, and samples (0.1-0.15 g) were snap-frozen in liquid nitrogen. Total proteins were extracted in protein extraction buffer [50 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM DTT, 2.5 mM NaF, 1.5 mM Na 3 VO 4 , 0.5% (w/v) Nonidet P-40, 50 mM β-glycerophosphate, and proteinase inhibitor cocktail] and centrifuged once at 6000 g, 4°C, for 20 min and twice at 17,000 g, 4°C for 10 min. Treatment with the calcium ionophore A23187. Twelve-day-old Col-0 seedlings were treated with 50 µM calcium ionophore A23187 (Sigma-Aldrich Co., MO, USA) for 15, 30, and 60 min. Samples were processed for the phosphorylation of MPK3 and MPK6 as described in the "Immunoblot analysis for detection of MPK3 and MPK6 phosphorylation" section. The leaf tissue was stored at −80°C until use.
Promoter-reporter imaging. The 3.0-kbp promoters for WRKY33 and CBP60g, both of which covered the previously analyzed respective regulatory sequences, were amplified from Col-0 genomic DNA by PCR and cloned into the pENTR/D-TOPO vector (Invitrogen). The promoter regions were recombined using Gateway technology into the binary vector pBGYN. The resulting pBGYN-pWRKY33-EYFP-NLS and pBGYN-pCBP60g-YFP-NLS vectors were introduced into A. tumefaciens GV3101 (pMP90) and then into Arabidopsis Col-0 plants using the floral dip method. A representative homozygous line was selected for each construct for further detailed analyses.
Promoter analysis. The statistical analysis for overrepresented transcriptional regulatory elements across transcriptome datasets described above was calculated using a prediction program 41 52,53 . To image the fluorescence from the GCaMP3 reporter (in Col-0 and gl1) in whole leaves, the adaxial sides of leaves from 4-week-old plants were brushed. To monitor the calcium waves propagating from trichomes, a single trichome from a 2-week-old seedling was flicked with a silver chloride wire. Samples were imaged with a motorized fluorescence stereomicroscope (SMZ-25; Nikon, Tokyo, Japan) equipped with a 1× objective lens (NA = 0.156, P2-SHR PLAN APO; Nikon) and an sCMOS camera (ORCA-Flash 4.0 V2; Hamamatsu Photonics, Shizuoka, Japan) 53 .
Propidium iodide staining. A stock solution of 10 mM propidium iodide (PI) was prepared with phosphate-buffered saline pH7.5. Rosette leaves of 4-week-old Col-0 plants were cut into 5 mm squares, floated in a glass petri dish with 20 µM PI solution, and incubated for 1 h at room temperature. Stained tissues were observed under the all-in-one fluorescence microscope (BZ-X800; Keyence Corporation, Osaka, Japan) equipped with a 20× objective lens (CFI S Plan Fluor LWD ADM 20XC, Nikon) and TRITC dichroic mirror (excitation 545/25 nm, extinction 605/ 70 nm) (KEYENCE).
Bacterial infection. MS was applied to the adaxial leaf surface of 4-week-old plants by brushing 4 times at an interval of 15 min for 3 h. Sample leaves were then inoculated by infiltration, using a plastic syringe (Terumo Tuberculin Syringe 1 mL; TERUMO), with Psm ES4326 (OD 600 = 0.001) resuspended in 10 mM MgCl 2 . Bacterial growth was measured 2 days after inoculation as described previously 78 .
Fungal infection. Alternaria brassicicola strain Ryo-1 was cultured on 3.9% (w/v) potato dextrose agar plates (PDA; Becton, Dickinson and Company, NJ, USA) for 4-20 days at 28°C in the dark. After incubation of the agar plates for 3-7 days under ultraviolet C light, a conidial suspension of A. brassicicola was obtained by mixing with RO water 79 . The adaxial side of leaves from 4-week-old plants was treated with 10 droplets or MS by brushing 4 times at an interval of 15 min for 3 h, followed by spotting with 5 µL conidia suspension (2 × 10 5 per mL) of A. brassicicola on the adaxial side of leaves. Inoculated plants were placed at 22°C under diurnal conditions (16-h-light/8-h-dark cycles) with 100% relative humidity. The lesion size of fungal infection was measured with ImageJ 3 days after inoculation.
Statistics and reproducibility. GraphPad Prism 9 (GraphPad Software, CA, USA) was used for all statistical analyses. One-sided or two-sided Tukey's multiple comparison test, one-sided Šídák's multiple comparison test, or two-way analysis of variance (two-way ANOVA) were used for multiple comparisons. In all graphs, asterisks indicate statistical significance tested by one-sided or two-sided Tukey's multiple comparison test, one-sided Šídák's multiple comparison test, or two-way ANOVA (multiple groups).