Spatiotemporal cytoskeleton organizations determine morphogenesis of multicellular trichomes in tomato

Plant trichomes originate from epidermal cell, forming protective structure from abiotic and biotic stresses. Different from the unicellular trichome in Arabidopsis, tomato trichomes are multicellular structure and can be classified into seven different types based on cell number, shape and the presence of glandular cells. Despite the importance of tomato trichomes in insect resistance, our understanding of the tomato trichome morphogenesis remains elusive. In this study, we quantitatively analyzed morphological traits of trichomes in tomato and further performed live imaging of cytoskeletons in stably transformed lines with actin and microtubule markers. At different developmental stages, two types of cytoskeletons exhibited distinct patterns in different trichome cells, ranging from transverse, spiral to longitudinal. This gradual transition of actin filament angle from basal to top cells could correlate with the spatial expansion mode in different cells. Further genetic screen for aberrant trichome morphology led to the discovery of a number of independent mutations in SCAR/WAVE and ARP2/3 complex, which resulted in actin bundling and distorted trichomes. Disruption of microtubules caused isotropic expansion while abolished actin filaments entirely inhibited axial extension of trichomes, indicating that microtubules and actin filaments may control distinct aspects of trichome cell expansion. Our results shed light on the roles of cytoskeletons in the formation of multicellular structure of tomato trichomes.

Introduction to have defective trichomes, indicating the involvement of microfilaments in morphology of tomato trichomes [33]. However our understanding of genetic and molecular mechanisms behind multicellular trichome formation in tomato remains fragmental.
In this study, we combined genetic screen with live cell imaging approaches to quantitatively study tomato trichome morphogenesis. We found both actin filaments and microtubules exhibited dramatically different patterns in different trichome cells in tomato. Such patterns also changed over developmental stages which could correlate with the distinct cell expansion. Genetic screen for aberrant trichome shapes confirmed the essential role of SCAR/WAVE and ARP2/3 complex in actin organization and trichome morphology. Microtubules appeared to define the anisotropy of the cell expansion while actin filaments seemed to play a major role in the extension of trichome cells. Taken together, our findings provide insights into the role of the cytoskeletons in multicellular trichome morphogenesis.

The morphology and organization of tomato trichomes
Unlike unicellular trichome in Arabidopsis, tomato trichomes include a total of seven types of morphologically and functionally distinct trichomes [7]. Among them, there are four types of glandular trichome (GT) and three types of non-glandular trichome (NGT) (Fig 1A and 1B). We first analyzed the abundance of each type of trichome in the stem of tomato (Micro-tom). We found that the type II and V represented two predominant types of NGT (S1A and S1B Fig).
The overall morphology of type II and type V NGT was similar but they consisted of different number of cells (Fig 1C and 1D). Majority of type II trichomes included five to eight cells, with eight cells as the highest ratio (27%) (Fig 1C). Type V trichomes displayed less varied cell number with three cells as the most predominant type (87%) (Fig 1D). The length and width of cells within type II trichome file became gradually reduced along the cell file (S1C-S1E Fig).
The average length of the largest basal cell was approximately 728μm which was seven times more than the smallest top cell. But the length of each cell within type V trichome file appeared to be similar despite all cells exhibited tapering shape (S1F-S1G Fig). Most examined type II trichomes still seemed to be in expanding as the cell length variation of each cell position within the cell file is high. In the basal cells, the longest cell was 1141μm in length while the shortest one was only 498μm (S1I Fig). Interestingly, the length of top cells shows wider range, with the longest and shortest ones as 298μm and 44μm respectively (S1I Fig). This suggested that the basal cells might be more mature and many top cells could be newly initiated. Based on the morphological survey, we decided to focus most of our analyses on type II trichome as this type with different cell numbers could represent typical expanding trichome stalk cells.

Visualization of actin organization at different developmental stages of tomato trichomes
Actin-based apparatus is also implicated in the morphogenesis of unicellular trichome in Arabidopsis [34]. To visualize how actin is spatially organized in different cells of tomato trichomes, we stably transformed tomato with the actin-binding marker Lifeact (p35S:Lifeact-eGFP). Among 18 transformed lines, 10 lines showed normal trichome morphology while the rest had certain level of defective trichome shape and abnormal actin bundles (S2 Fig). The ten normal lines were then chosen for further live-cell imaging analyses. To visualize different developmental stages of type II trichomes, we imaged young leaves (14 days after germination) and identified trichomes as early as one-cell and initiating stage (Fig 1F-1K). At one-cell stage, up to 86% actin filament showed transverse (0-15 degree against the horizontal plane) or oblique  degree) alignment, and the longitudinal actin filaments were only 14% (Fig 1F and 1L). However, the actin alignment in this cell seemed to be spatially heterogeneous, with transverse actin filaments girdled at the conjunction base and longitudinal actin aligned beneath the protruding dome ( Fig 1F).
As the cell number increased, type II trichomes exhibited distinct actin patterns in different cells. The basal cell maintained around 80% transverse or oblique actin filaments while these alignments gradually reduced and longitudinal actin filaments progressively increased upward along the trichome (Fig 1F-1L). At 6-cell stage, longitudinal actin filaments became over 60% in distal top cells ( Fig 1L). These observations suggested that actin filaments rearranged during the elongation of tomato trichomes. Interestingly, almost all trichome cells including the initiating one-cell stage one appeared to be highly vacuolized during all developmental stages as most actin tended to assemble into cortical network (Fig 1F-1I).
When the trichome reached six-cell stage, most cells further elongated and actin filaments exhibited the overall spiral arrangement along the long axis (Fig 2A). We next quantitatively measured the orientation using FibrilTool (Fig 2P). Interestingly, the extent of helix appeared to be progressively reduced from the basal cell to the distal cells at the top. In the first basal cell, actin filaments aligned approximately 34 degree against the direction perpendicular to the axis of the growth, while this angle gradually increased along the cell file and eventually formed the longitudinally actin bundles (Fig 2B-2G and 2Q). Similar to type II trichomes, actin filaments exhibited spiral-to-longitudinal pattern in type V trichomes (S3 Fig).
The longitudinal actin arrangement in the distal top cells physically resembled the actin bundles in the shank of tip growth cells such as pollen tubes and root hairs. However, actin filaments often formed a doom-shaped actin cap at the protruding head of the top cells, suggesting it might be a different mechanism from the typical tip-growth (Fig 2M and 2N). Within most cells, the actin arranged into transversely aligned mesh on basal and apical sides, while actin in the middle part of the cell showed more helical alignment ( Fig 2H). Similar to the younger stages, most six-stage trichome cells were highly vacuolized as evidenced by the cortical actin organization (Fig 2I-2L and 2O). It is possible that this non-uniform organization of actin within a cell reflect the putative distinct expansion rate of the cell surface.

The dynamics of actin filament in different trichome cells
To understand how distinct actin patterns form in different trichome cells, we performed timelapse imaging in trichomes at initiation, one-cell and four-cell stages (Fig 3 and S1-S3 Videos). When a trichome started to initiate from epidermis, the nuclear moved into the bulging site and actin filament formed a cap-like mesh under the protruding doom (Fig 3A-3F and S1 Video). In the upper part of the cell, almost all actin filaments in the cytoplasm appeared to be longitudinal (Fig 3E and S1 Video). At the one-cell stage, we found that the actin filaments were rarely depolymerized and often moved up and down or left and right (Fig 3G-3J). Meanwhile, longitudinal actin filaments sometimes formed transverse or oblique branches ( Fig 3I). As the four-cell type II trichomes represented the intermediate stage of the trichome development (Fig 3K-3Q), we thus chose this stage for further dynamics analyses. In top cells, we observed high frequent Roles for actin filaments and microtubules in forming multicellular structure of tomato trichomes depolymerization of transverse actin filament (Fig 3L and 3N and S3 Video). In contrast, longitudinal or oblique actin filaments tend to polymerize or only shifted left and right (Fig 3M, 3N and 3P). In the basal cell, transverse actin filaments were dominant and most transversely aligned actin appeared to be less dynamic ( Fig 3O). In this cell, oblique actin filaments tended to have higher frequency of polymerization to form actin cable ( Fig 3O). These observations indicated distinct actin patterns in different trichome cells formed through regulating the dynamics of differentially oriented actin filaments.

Organization and dynamics of microtubules in tomato trichomes
During the formation of unicellular trichome in Arabidopsis, microtubules are mostly involved in trichome branching [6,35]. To understand how microtubules regulate trichomes in tomato, we performed live imaging of microtubule organization by visualizing stably transformed  We first looked the microtubule organization in trichomes at different developmental stages. At one-cell stage, EB1a-eGFP showed as dots spread at the tip of the cell, and the EB1a-GFP seemed to cover the whole dome of trichomes (Fig 4A-4C). In the basal part of the cell, EB1a-eGFP displayed a comet shape, a sign of rapid moving track during the confocal scanning (Fig 4A-4C). By tracking the comet shape, we could infer the direction of microtubule movement. In the basal part of cells, many EB1a-GFP appeared to traffic spirally (Fig 4D). At two-cell stage, there were also two groups of EB1a signal, with less dynamic ones on the membrane of protruding dome and quickly moving ones in the basal cell of the trichome (Fig 4E and 4F). The time-lapse imaging showed that EB1a-GFP frequently moved to the tip of the trichomes (Fig 4I and 4J and S4 Video). In the basal cell, we could easily detect the comet shaped EB1a-eGFP (Fig 4E-4H). Further live time-lapse imaging in the basal cell of the three-cell stage showed that the comet EB1a-GFP moved spirally up or spirally down (Fig 4K and 4L). At the six-cell stage, EB1a in trichome cell files showed an overall helical movement pattern (Fig 4M and 4W). Further immunofluorescence imaging using anti-TUB antibodies in observed cells (many other cells were lost due to the specimen processing) displayed the similar spiral pattern as shown by EB1a-eGFP (Figs 4S and S4G). Interestingly, the movement pattern of EB1a displayed a similar angel transition as actin filaments (Fig 4N-4W).

dt mutants distinctly affect different trichomes cells
To uncover the genetic and molecular mechanisms controlling the cell expansion of tomato trichomes, we examined EMS mutagenized lines. We identified six mutants with visibly aberrant trichome morphology and named them as distorted trichomes-1~6 (dt-1~6) (Fig 5A-5D; dt-4~6 were not displayed). Compared with wild type in which trichomes were straight and perpendicular to the epidermal surface, the trichomes on stems and leaves of all mutants were curly and prostrated under stereomicroscope (Fig 5E-5L). The phenotypes of mutant trichomes were more prominent under scanning electron microscopy (SEM). The trichomes of WT were a group of cells with tapered diameter that were connected end to end. However, the first basal cell of type II trichomes in mutants became dramatically swollen in one direction while the middle cells bent with little swollen, and the top cell curved into a hook shape (Figs 5M-5X and 6A and 6B).
To better understand the morphological defects of dt trichomes, we quantitatively measured the geometry of type II trichomes in both WT and dt mutants. In the mutants, all trichome cells became shorter and wider, with the most pronounced defect in the first basal cell. The average length of first basal cells in mutants was about 150μm, about one-seventh of that in WT. However, the mean width was about 180μm in mutants which was two times than the WT. The extent of morphological phenotypes in mutant trichomes gradually reduced along the cell file, and the top cell was only half of the WT length while the width was higher than WT ( Fig 6F and 6H). Compared to the type II trichomes, the change of overall morphology of type V trichomes was marginal, with cells becoming shorter but much curved (Fig 6D-6E). Unlike type II trichome, the overall morphology of type V was only slightly changed, without visible swollen cells. Furthermore, the phenotype of type III trichomes in dt mutants was similar to that observed in type II, with swollen and curved cells (Fig 5Q-5T). Interestingly, the morphological change of glandular trichomes was more like type V non-glandular trichomes, with the stalk cells bending. The glandular cells locating on the top of the type VI trichomes seemed to be WT-like shape and diameter, suggesting the morphology of these cells was independent of the mutant genes (   Roles for actin filaments and microtubules in forming multicellular structure of tomato trichomes

Genes defective in dt mutants encode WAVE-ARP2/3 homologs
To identify causative mutations, we developed six F 2 mapping populations by crossing mutants to the wild type. The trichomes of all F 1 plants were normal and the separation ratio of F 2 population was consistent with Mendel's separation law (3:1) (S1 Table). This result indicated that the distorted trichomes phenotype was regulated by recessive genes. Using Nextgeneration sequencing-based bulked-segregant analysis (BSA) approach, we identified an associated locus on chromosome 9, including two candidate genes, Solyc09g014980 and Solyc09g065130 (Fig 7A). According to the tomato reference genome sequence (https://www. solgenomics.net/), Solyc09g014980 encodes a Wiskott-Aldrich syndrome protein family member and is homology to the dis3 (AtSCAR2) (Fig 7D). Similar as AtSCAR2, SlSCAR2 consists of four conserved domains, AHD, B, WH2 and A domain [36]. The allele of AtSCAR2 in dt-1 contained one nucleotide mutation (5128C-T), resulting in an early stop codon and deletion of A domain that is responsible for activating ARP2/3 complex. PCR amplification followed by  Roles for actin filaments and microtubules in forming multicellular structure of tomato trichomes length SlSCAR2 driven by CaMV35S promoter. The trichomes on stems and leaves of transgenic plants (28 out of 30 of T0 plants) were rescued by the full-length SlSCAR2 fragment ( Fig  7G-7J). Secondly, we knocked out SlSCAR2 in WT by CRISPR/Cas9 and further confirmed the mutation sites by DNA sequencing (S8 Fig). As shown in S4 Fig, we isolated three CRISPR/Cas9 mutant lines. Compared with WT, the trichomes on stems and leaves of all three CRISPR/Cas9 mutants became distorted (Fig 7K-7N).
Using the same strategy, we cloned the mutated genes from other trichome mutants (dt 1-6). There are ten candidate genes from BSA analysis of the F2 population of dt-2 and WT cross. One nucleotide mutation (211A-T) of SlARPC4 resulted in an early stop codon in dt-2 (Fig 7B and 7E). In dt-3, we identified an amino acid substitute (223 G to D) in SlARPC1B. Finally, we identified two candidate genes of ARP2/3 complex and three SCAR/WAVE complex homologs from six dt mutants (S2 Table). Two among the five genes we identified (SRA1 and ARPC2A) were also reported previously [33,37]. As genes identified through our forward genetic screens all belong to the complexes regulating actin filaments, this type of cytoskeletons likely represents an indispensable regulator of trichome morphogenesis.

The spatial configuration of Actin filaments is altered in dt mutants
To visualize actin filament organization in dt muatnts, we introduced 35S:Lifeact-eGFP into the dt-1 through stable transformation. Compared with WT, the orientation of actin filament in dt-1 lost the helix pattern in all trichome cells, and instead formed longitudinal or random actin bundles. Transversely oriented cortical actin meshes in basal and apical region were only observed in some cells (Fig 8A and 8B). In the first basal trichome cell in dt-1, the actin often arranged into a girdling belt-like structure that was perpendicular to the growth direction on the most swollen side of the cell. Interestingly, on the other side of the most swollen part, the actin filaments appeared to be diminished (Fig 8B-8D). When visualized in the cross section, the actin appeared to organize not only in the cortical region but also in the cytoplasm, and the amount of cytoplasmic actin cable was significantly increased (Figs 8E and 8J and S9C). In the bending cells, longitudinal actin filaments gathered into cables along the adaxial side of the cell, leading to the aberrantly aggregated actin cables (Figs 8H and 8I and S9A and S9B). Compared with WT, anisotropy of actin filament was significantly reduced in the basal and top cell (Fig 8J), which was consistent with the severe cell morphology.

Microtubules are essential for actin organization in tomato trichomes
To functionally dissect the roles of microtubules and actin filaments in tomato trichomes, we treated the tomato seedlings with 20μm/L Oryzalin and 20μm/L Latrunculin B for four days respectively (Fig 9A). Both disrupted microtubules and actin filaments remarkably inhibited trichome morphology (Fig 9A, 9C, 9F and 9J). With Oryzalin treatment, all trichome cells became radially swollen, resembling the isotropic expansion of microtubule-disrupted root cells (Fig 9G and 9H). When actin configuration was abolished, the axial extension of trichome cells was significantly repressed while the overall anisotropy seemed to be retained (Fig 9K and  9L). Interestingly, Oryzalin treated cells with swollen shape exhibited increased bundling of actin cables and reduced fine actin filaments (Figs 9I and S9D). In addition, the thick actin cables appeared to align more transversely, possibly girdling the expansion force. These results suggest that microtubules presumably function as a major player to maintain the cell expansion anisotropy while actin filaments are more important to promote the axial elongation of the cell in trichomes. We further quantified the trichome initiation using SEM images. Despite actin filaments were all removed by the treatment, trichome initiation was only slightly affected. Similarly, disassembly of microtubules seemed to only influence the anisotropy of cell expansion, without affecting the trichome initiation rate (Fig 9B). Thus cytoskeletons possibly control tomato trichomes mainly through morphological regulation.

Discussion
Both microtubules and actin filaments have been shown to promote morphogenesis of unicellular trichome in Arabidopsis. Previous cell imaging and genetics evidence suggest that microtubules are more involved in trichome branching in Arabidopsis [29][30][31]. However, the regulation in multicellular tomato trichomes appeared to be more complex. Microtubules seemed to control expansion anisotropy which is similar to its canonical roles in many other cell types [38]. The tomato trichome initiated with the nuclear moving into the bulging site, and both microtubules and actin filaments formed longitudinal network to promote the protrusion from epidermis ( Fig 10A). The spiral pattern and gradual transition from transverse to longitudinal actin alignment presumably represent the different stage of trichome expansion. Furthermore, the cell length of the top cell is more variable than the basal cell. The basal cells could reflect the more fully expanded status while the rest cells are still in expanding. Since most cells within a trichome need to increase both length and width, it is reasonable to infer that most trichome cells undergo diffuse expansion. However, compared to the canonical cytoskeleton organization in diffusely expanding cells, both actin filaments and microtubules appear to organize in dissimilar manner in different trichome cells in tomato. During expansion, root cells form transverse cortical microtubules and longitudinal actin filaments while leaf pavement cells have actin aggregated in the lobe protruding region and microtubules bundled at the indentation [39]. In tomato trichomes, transverse alignment of both actin and microtubules are restricted to fully expanded basal cells or two ends of an expanding cell. In the middle part between the apical and basal end of a trichome cell, both cytoskeletons form spiral organization which could be important for promoting the cone-shape. In addition, actin filaments organize in the similar helix as microtubules, suggesting a potential coordination between the two cytoskeletons ( Fig 10B). When microtubule was disrupted by oryzalin, actin filaments bundled into cables and lost longitudinal alignment in trichome cells, providing further support of the interaction between the two cytoskeletons.
Although the paradigm of cytoskeletons in cell expansion is directing the arrangement of cell wall components, actin filaments in trichomes also seemed to respond to mechanics derived from cell expansion [13,40,41]. The spiral pattern of actin filaments in most cells between the basal one and the distal top one presumably results from the reciprocal interaction between cytoskeletons and cell wall. The spatial distribution of cytoskeletons controlled by mechanical heterogeneity within a cell was previously reported in a number of cellular systems [5,42] However, the correlation between cytoskeletons and mechanical force seems to be dissimilar in different systems. In epidermal pavement cells, the microtubules were often found to organize in bundles parallel to the direction of maximal tensile stress [3,43]. During trichome branching in Arabidopsis, microtubules reorient circumferentially in the collar region of the emerging site [29]. In tomato trichomes, both microtubules and actin filaments gradually re-orient from transverse array to helical organization, and eventually to longitudinal alignment in top cells. This pattern possibly reflects the mechanical distinction derived from cell expansion in those cells. Besides, the changing direction of cytoskeletons in different cells could associate to different expansion stages of trichome cells.
In addition, the actin filaments in dt mutants re-aligned into the girdling structure along the maximal mechanical forces caused by swollen trichome cells. Intriguingly, we observed a large number of actin bundles formed across the cytoplasm connecting expanding cell walls on different sides (Fig 10C). These observations provide indirect evidence supporting actin filaments interplay with mechanical stress within a cell. During expansion, the cell wall exhibits a great extent of elasticity through constantly loosening and stiffening. Thus the possible scenario is cytoskeletons not only provide a mechanical constraint to the cell wall elasticity, but alter their own alignments in response to the tensile stress, giving a feedback to amplify the outgrowth signal and eventually shape the cell.
It was observed in Arabidopsis trichome that the early bulging epidermis has few cortical microtubules while contains prominent actin meshwork [44]. By contrast, microtubules seemed to play a major role in trichome branching [30,31]. The apical zone of the trichome branching always contains a microtubule-depleted zone while microtubules form transverse bundles at the base of protruding tip [28]. In contrast, live-imaging of EB1a marker showed that there was EB1a-GFP signal in the apex of the top trichome cells (Fig  4B and 4F). However, whether there is a microtubule configuration in protruding domain of tomato trichomes remains to be further verified using more direct markers including GFP-Tubulin or GFP-MBD. We also observed longitudinal cytoskeleton bundles of bulging site of epidermis, which suggest that actin filament and microtubule cytoskeletons may play distinct roles in polar growth of the tomato trichome. In the elongating tomato trichome cells, both cytoskeletons were either longitudinally or spirally aligned along the growth axis, with the transverse organization visualized only in basal cells or apical and basal ends of other elongating cells. However, this pattern is not specific to multicellular structures as it was recently observed that unicellular cotton fiber also has a longitudinal actin filament with transversely aligned microtubules [32]. When tomato trichomes were treated by oryzalin, all trichomes became isotropically expanded balloon like structure, indicating microtubules play canonical roles in cell expansion. By contrast, disruption of actin filaments by latrunculin B produced short trichomes with normal anisotropy, suggesting actin filaments are involved more in cell elongation. Interestingly, pharmacologically disrupted either cytoskeletons failed to inhibit the trichome outgrowth from epidermis. In addition, we observed the nuclear movement into the bulging site before the significant outgrowth in epidermis, suggesting the bulging process is likely regulated by a cytoskeleton independent mechanism.
The major role of actin in trichome cell elongation can be further supported by the observation of predominant arrangement of actin along the extension axis in the top cells, resembling the actin structures in tip-growing cells. In tip growth, actin bundles, rather than microtubules function to direct the deposition of cell wall materials via secretion [45]. The actin cables determine the location and direction of cell growth by targeted secretion of vesicles containing cell wall materials transporting along actin filament. In addition, actin cables could also provide the mechanical force for the outgrowth as suggested in pavement cell lobes and trichome initiation in Arabidopsis. The longitudinal alignment of actin in top cells of trichomes could drive the directional vesicle transport and provide leading force for the cell elongation. The location and direction of the cell protrusion seemed to reply mainly on proper actin configuration as loss of SCAR/WAVE and ARP2/3 complex led to substantially curving cells. In dt mutants in which actin filaments became defective and bundled, most cells except for the basal one became distorted and curved but maintained the overall expansion anisotropy. Together with the pharmacological assay of actin filaments in trichome morphology, our observations suggest that microtubules and actin function on different aspects of the trichome cell morphology with microtubules more on expansion anisotropy and actin more on trichome elongation.

Morphological and cellular analysis
The abundance analysis of trichomes was conducted on the stem between third and fourth leaves of 30 days seedlings using stereomicroscope (Nikon, SMZ18). More than 6 independent WT plants and 4 independent mutant plants were measured. For quantification of the geometric parameters of trichomes, images of trichomes was taken by DIC (Nikon, ECLIPSE Ni-U) and cell number of each trichomes, cell length and cell width was manually measured using Image J software. More than 200 trichomes from 10 independent WT plants and 30 trichomes from 6 independent mutant plants were measured.

Construction of marker lines
The Lifeact, a 17-amino-acid peptide that is derived from the yeast actin binding protein ABP-140 [46], is regarded as being the most authentic marker for actin organization and distribution. Microtubule end-Binding protein EB1 binds to all growing MT plus-ends. Lifeact and EB1a (At3g47690), fused with eGFP respectively, were introduced into the pHellsgate 8 vector under control of the CaMV 35S promoter. The construction of 35S::Lifeact-eGFP and 35S:: EB1-eGFP were transformed into Micro-Tom tomatoes by Agrobacterium-mediated genetic transformation.

Quantification methods
For the quantification of the geometric parameters of trichome cells, cell heights, cell widths and actin angles were manually measured using ImageJ software. More than 200 trichome cells from 10 independent wild-type plants and 30 trichome cells from 6 independent mutant plants were measured. FibrilTool, an ImageJ plug-in, was used for quantification of the orientation angle and the anisotropy of actin arrays in a given region of interest of the wild type and dt-1. Anisotropy values range from 0 to 1. 0 indicates pure isotropy, and 1 represents pure anisotropy. Statistical analyses were performed using standard's t test. Data were represented as the mean ± SE from at least three independent experiments. Not significant P > 0.05, 0.01 < � P < 0.05, �� P < 0.01.

Confocal laser scanning microscopy
To visualize actin and microtubules, trichomes of transgenic plants stably expressing Lifeact-eGFP or EB1-eGFP were imaged by a Zeiss LSM 880 confocal laser scanning microscope with a 40× water lens. For imaging 3D reconstruction of trichomes, Z stacks of confocal images from the distal regions of trichomes with stably expressing green fluorescence were taken from the top view along the Z axis at steps of 0.8 μm, and were used to reconstruct the 3D images using Zeiss LSM 880 software.

Mapping populations
Mapping populations were constructed by crossing WT (Micro-Tom, female) and trichomes mutant (male). F2 population obtained by selfing the F1. To determine the mutation is whether mono-gene recessive or not, F2 segregation analysis was conducted.

Bulked segregant analysis (BSA)
Genomic DNA of 30 individuals clearly exhibiting mutant-like phenotypes from F 2 population and 30 WT individuals was extracted by CTAB. All DNA quality and concentration were checked and then mixed to construct two bulks (mutant-like bulk; WT bulk). Each bulk was sequenced to a depth of 30× coverage of the tomato genome by HiseqXten-PE150 (Novogene).
Trimmed sequences are mapped onto the tomato reference genome (Heinz 1706 cultivar) and EMS mutation variants are filtered. Analysis of the allelic variant frequencies in the pools leads to the identification of the causal mutation with very high frequency in the mutant-like bulk (ideally 100% having the variant allele). The genes with the expected allelic frequency tend to be 1 for the mutant-like bulk was performed mutation identification and transgenic verification [47]. Then, the candidate genes were cloned and sequenced to verify the mutant site. The primers were listed in the S3 Table.

Transgenic analysis
To verify the candidate genes, two expression vectors were constructed. The full-length coding sequence of candidate genes from WT was inserted into the pHellsgate 8 vector under control of the CaMV 35S promoter. For construction of CRISPR vector, target sequence was designed based on CRISPR-PLANT (https://www.genome.arizona.edu/crispr/). Then, the target sequences were introduced into the pTX vector in which the target sequence was driven by the tomato U6 promoter and Cas9 by 2×35S. The recombinant pHellsgate8 vectors were transformed into corresponding mutants and CRISPR vectors were transformed into WT, mediated by Agrobacterium tumefaciens strain C58. The primers were listed in the S3 Table.

Latrunculin B (Lat B) and Oryzalin (Ory) treatment
Freshly sprouted tomato seeds were treated on the 1/2 MS medium with 20mmol/L Lat B and 20mmol/L Ory for four days. Controls and treatments were grown in the growth chamber with a 16h light/8h dark cycle, 22˚C temperature. Phenotypes were imaged using DIC (Nikon, ECLIPSE Ni-U), scanning electron microscopic (SEM) and Zeiss LSM 880 confocal laser scanning microscope.

Antibodies and Immuno-staining
For immunostaining, trichomes with sliced leaves were fixed for 1h at room temperature with 4% paraformaldehyde in 50 mM PIPES, pH 6.8, 5 mM EGTA, pH7.0, 2 mM MgCl 2 , and 0.4% Triton X-100. The fixative was washed away with PBS buffer, and samples were treated for 15 mins at room temperature with the solution of 0.1% Cellulase R-10 and 0.1% Pectinase Y-23. And then, samples were incubated for 30 min in 1% (w/v) BSA in PBS and incubated for 1 h with monoclonal anti-α-tubulin-fluorescein isothiocyanate antibody produced in mouse (1:2000, Sigma F2168) at 37˚C. Samples were then washed three times for 10 mins in PBS and incubated for 1 h with secondary antibodies (1:500) at 37˚C. After washing in the PBS buffer with 0.1% p-Phenylenediamine, 0.1 M Propyl gallate, 50% Glycerol, pH9.5, samples were imaged using the Zeiss LSM880 confocal microscopy.