Fluorescent cytoskeletal markers reveal associations between the actin and microtubule cytoskeleton in rice cells

Rice (Oryza sativa) is one of the staple food crops for billions of people around the world (Seck et al., 2012). Understanding the response mechanisms of rice to diverse intra- and extra-cellular signals is tightly linked to improvements of rice performance. The plant cytoskeleton, composed of filamentous actin and microtubules, critically supports plant development and stress resistance in plants (Kost et al., 1999; Wang and Mao, 2019), including rice. Although most plant cytoskeleton studies have been performed in the model dicot plant Arabidopsis thaliana, there are also a few studies that highlight the importance of the cytoskeleton in rice development and environmental responses. For example, both actin- and microtubule-related rice mutants display severe developmental defects (Komorisono et al., 2005; Zhang et al., 2010), microtubule-related proteins were frequently reported to control rice seed size (Sun et al., 2017; Yang et al., 2020), and several rice cytoskeleton proteins were found to impact stress resistance (Xu et al., 2018; Lo et al., 2020). Therefore, it is important to study the cytoskeleton in detail in rice.

Most of the rice cytoskeleton reports focus on mutant characterization, with negligible in vivo cytoskeleton investigations (Zhang et al., 2010; Li et al., 2014; Huang et al., 2018; Song et al., 2019; Lo et al., 2020; Chang et al., 2021). In particular, important processes in rice, including tillering, underwater growth, pollen development and seed quality, have not yet been studied in connection with the cytoskeleton despite its prominent roles in these processes. Whereas static studies by immunolabelling are dominating the rice cytoskeleton field, live-cell imaging is superior because the cytoskeleton is composed of highly dynamic structures (Shaw et al., 2003; Sheahan et al., 2004). To this end, bright and stable fluorescent markers of the cytoskeleton are much needed. However, fluorescent cytoskeleton marker lines have only been established in some plant species, e.g. Arabidopsis, Medicago (Medicago truncatula), maize (Zea mays) and cotton (Gossypium spp.) (Shaw et al., 2003; Timmers et al., 2007; Mohanty et al., 2009; Yu et al., 2019). These studies have highlighted the importance of selecting suitable fluorescent tags and cytoskeleton-labelling proteins to minimize their influence on cytoskeletal behaviour and allow long-term visualization of the cytoskeletal structures (Sheahan et al., 2004). The fine Actin-Binding Domain 2 of Arabidopsis fimbrin 1 (fABD2), mTalin and Lifeact are commonly used for actin labelling (Kost et al., 1998; Sheahan et al., 2004; Qu et al., 2017), while tubulin subunits and the mammalian Microtubule-Associated Protein 4 (MAP4) are typical proteins for microtubule labelling (Marc et al., 1998; Ueda et al., 1999; Shaw et al., 2003). Notably, these labelling proteins may impact the behaviour of the cytoskeletons, and so the protein abundance and biochemical properties of marker proteins are important factors to consider (Courtemanche et al., 2016). However, the impacts of the marker proteins appear to differ among different organisms. That may be the reason why one construct performs well in one species or tissue, but not in another (Vidali et al., 2009; Flores et al., 2019), emphasizing the need to choose suitable components to construct markers for specific plants or tissues.

In this study, we generated effective rice fluorescent cytoskeleton marker lines, including those of filamentous actin and microtubules, allowing live-cell imaging of the rice cytoskeleton. With our marker lines, we explored several cytoskeleton-related topics in rice biology, including actin and microtubule interplay during pollen development and root cell division, salt response and cytoskeletal mutant characterization. We thus describe several exciting phenomena, which will open up further investigations on rice cytoskeleton functions.

To study cytoskeletal dynamics in rice in vivo, we set out to generate fluorescent marker lines for the microtubule and actin cytoskeleton. We generated a range of different combinations of promoters, fluorescent proteins and cytoskeleton-labelling proteins (Fig. 1). We transformed these into wild-type rice plants and screened multiple independent transgenic lines in T2 generation. The advantages and disadvantages of these tested components are compared and summarized in Table S1. We found that the monocot ubiquitin promoters including rice ubiquitin 2 (RUBQ) and maize ubiquitin 1 (Ubi1) gave homogeneous fluorescent signals in rice cells, while the tobacco virus promoter 35S was often silenced (Fig. 1A,B). RUBQ lines typically showed brighter signals than Ubi1-driven lines. Inspired by Arabidopsis studies, we tried the domain of mouse MAP4 and the rice Tubulin subunit Alpha 1 (OsTUA1) to label rice microtubules, and Arabidopsis actin-binding protein fABD and its rice homolog OsfABD to label rice actin filaments. In the resulting transgenic lines, OsTUA1 labelled rice microtubules weakly, but substantial fluorescence was also observed in the cytosol (see white arrowheads in Fig. 1B). No signal was detected in the OsfABD lines (second image in Fig. 1A). Fortunately, MAP4 and fABD lines effectively labelled the microtubule and actin cytoskeleton in rice plants (right three columns in Fig. 1A,B). Notably, fABD lines were developmentally indistinguishable from wild-type plants, whereas the MAP4 lines occasionally caused some growth defects also reported for different microtubule marker lines in Arabidopsis (Fig. S1A). However, wild-type-like MAP4 transgenic lines were also frequently obtained (Fig. S1B). Among the fluorescent proteins we used, the bright green fluorescent protein mNeongreen (mNG) appeared bright and stable, while the red fluorescent protein mScarlet was bright but prone to photo-bleaching (Fig. S1C,D).

We first visualized the cytoskeleton throughout different rice tissues, including roots, root hairs, stem, leaves, stomata, anthers, pollen grains and pollen tubes (Fig. 2A,B). Bright fluorescence was observed in all tissues, except in the pollen mother cell before microspore formation (Fig. S2). Overall, we observed brighter fluorescence in younger plant tissues. Thus, our marker lines can be used to image the organization and dynamic behaviour of the cytoskeleton in both vegetative and reproductive tissues (Fig. 2C,D, Movies 1, 2), including cells with different growth modes.

Using our marker lines, we set out to explore a range of processes related to the cytoskeletons. We first imaged cytoskeletal behaviour during rice pollen development. As shown in Fig. 3, both the actin and microtubule cytoskeletons were fragmented in young microspores (left column). Short actin filaments and microtubules subsequently formed during pollen grain development (bright-field panels from left to right in Fig. 3). Both actin filaments and microtubules gradually extended as the pollen grains increased in size; however, the polymers lacked clear common directionality. During pollen grain maturation and amyloplast formation (red asterisks in Fig. 3), both the actin and microtubule cytoskeletons formed radial patterns radiating from the germination pore (indicated by yellow arrows in Fig. 3). Once the pollen was hydrated, the cytoskeleton gradually became disordered again and partially depolymerized (right column in Fig. 3).

The radial alignment of the actin and microtubule cytoskeletons prompted us to consider whether the two structures perhaps function together and associate with each other. Therefore, we obtained actin and microtubule cytoskeleton dual-labelled lines (RUBQ::mNG-fABD and RUBQ::mScarlet-MAP4) by crossing the single marker lines. In the resulting lines, we observed bright fluorescence from both markers and subsequently observed actin and microtubule cytoskeletons simultaneously in mature pollen grains. Strikingly, we found that the two fluorescent signals overlapped (Fig. 4A,B). Indeed, the two markers showed 70% colocalization based on Pearson correlation coefficient analyses (Fig. 4C). Closer observations revealed that thick actin bundles radiated out from the germination pores (indicated by the white arrowheads in Fig. 4D), whereas filamentous actin interspersed these bundles (indicated by the yellow arrowheads in Fig. 4D). Interestingly, the colocalization of actin and microtubule cytoskeleton occurred along the actin bundles, and not the fine filaments (as shown in Fig. 4D and Movie 3). This is consistent with colocalization analyses using Mander’s coefficient test, in which up to 90% of the microtubule-related fluorescence colocalized with the fluorescence from the actin cytoskeleton, whereas only ∼50% of the actin-related fluorescence colocalized with the microtubule signal (Fig. 4C).

To address whether the two cytoskeletal structures are associated with each other in mature pollen grains, we recorded time-lapse imaging of the dual-labelled line. Here, we found that the microtubules were coordinated with the actin bundles (Fig. 4E,F), indicating a tight interplay between the two cytoskeletal structures. This indicated that there may be a mutual structural and perhaps a functional relationship between the cytoskeletal components in pollen grains. We investigated this by disrupting one of the polymers and observed whether this resulted in any changes in the other one. When the actin cytoskeleton was partially depolymerized by actin inhibitor latrunculin B, the number of microtubule polymers did not obviously change (Fig. 4G, Movie 4). However, the microtubule radial organization was perturbed and the engagement with actin decreased, i.e. the Pearson correlation coefficient between the microtubules and the remaining actin filaments was reduced (Fig. 4G, Fig. S3). Notably, when the microtubules were depolymerized using oryzalin, the radial actin bundles rapidly disappeared and were replaced by a myriad of cytosolic actin filaments (Fig. 4H, Movie 5). Overall, the oryzalin treatment was accompanied by a reduction in actin filamentous fluorescence (middle column in Fig. 4H). These observations indicate that the actin and microtubule cytoskeleton provide mutual structural support to each other during pollen grain development.

Motivated by the cytoskeletal interactions in rice pollen, we next examined their relationship in other cell types, such as root cells. Here, microtubules play an essential role during cell division and display distinct configurations at different division stages, including preprophase band, spindle and phragmoplast (Lipka et al., 2015). These configurations were also clearly observed in dividing rice root cells by our marker lines (Fig. 5). Actin filaments function in cell division (Maeda et al., 2020); however, the actin configurations during cell division remain controversial (Panteris, 2008). In rice root cells, the abundant actin filaments in non-dividing cells gradually disappeared during preprophase band formation (Fig. 5). At telophase, we detected scant actin cytoskeletal fluorescence (Fig. 5). However, we noted clear actin-related fluorescence during phragmoplast progression (indicated by the arrowhead in Fig. 5), where the fluorescence appeared to be associated with the emerging cell plate. This observation is consistent with the reported cell plate-promoting function of actin cytoskeleton (Maeda and Higaki, 2021). Hence, in contrast to in the pollen grains, the actin and microtubule cytoskeletons appear to function separately in dividing rice root cells. The disappearance of the actin cytoskeleton may coincide with reduced cytoplasmic streaming, which decreases during the onset of division in Paramecium bursaria (Sikora et al., 1991).

In interphase rice root cells, we did not detect any obvious association between the actin and microtubule cytoskeletons (Fig. 6A). As reported in Arabidopsis, transverse microtubule arrays in elongating root cells re-arrange towards a longitudinal direction during cell maturation and growth cessation (Chan, 2012; Adamowski et al., 2019). We also noted a similar re-orientation in developing rice roots (top row in Fig. 6A). Interestingly, we also found a similar re-organization of the actin cytoskeleton (bottom row in Fig. 6A). To quantify such change, we extracted filament edge angles using the Cytoseg2.0 algorithm (Nowak et al., 2020) and plotted the angles as circular histograms (Fig. 6B). We analysed the perpendicularity (i.e. deviations from the cell growth axis of the filaments) and randomness (i.e. the degree of coherence of the filaments) of the resulting histograms (Fig. 6C,D). In agreement with our observations, both the actin and microtubule cytoskeletal arrays re-organized to align with the growth axis of the maturing cells (Fig. 6C). Whereas the actin array became increasingly more directional in maturing root cells, the microtubule array re-oriented in a coordinated fashion where the array maintained high directionality during the re-arrangement (Fig. 6C,D). We also noted similar actin array re-organization in other maturing cell types, such as anther epidermis (Fig. S4). We speculate that the cytoskeletal array re-organization may be a common procedure during interphase cell maturation.

Apart from supporting plant development, the cytoskeleton also plays crucial roles in adapting to changes in environmental conditions. For example, the microtubule array is sensitive to salt (NaCl) treatment, which causes rapid depolymerization in Arabidopsis (Endler et al., 2015). To assess how salt impacts the cytoskeletons in rice, dual-labelled rice seedlings (5 days after the crown root emerged) were incubated in NaCl-containing (250 mM) liquid medium. As in Arabidopsis, microtubules rapidly de-polymerized and re-emerged over time (Fig. 7A, Table S2). The response period and extent correlated with seedling age and root cell type. The younger the seedling, the more rapidly it responded (Fig. S5). Microtubule adjustments, both de-polymerization and recovery, started from the root tip and progressively responded upwards along the root (Fig. S6). During microtubule de-polymerization, we observed apparent fluorescent foci. These were initially few and relatively small, but gradually grew larger with time until microtubules disappeared (Fig. 7A from 2 h to 4 h, Table 1). During microtubule recovery, the large foci seemed to form new microtubules and we only observed small foci that gradually disappeared (Fig. 7A from 6 h to 8 h, Table 1). At later time points, microtubule arrays recovered and almost no foci were observed (Fig. 7A 8 h, Table 1, Table S2). However, the microtubule density never reached the levels of those prior to treatment (Fig. 7A, control versus 8 h, Table S2), 77.1 microtubule filaments per cell (counted from ten control cells) compared with 13.79 in cells treated by salt for 8 h, indicating that microtubule recovery was only partial. In contrast to the microtubule array, no major depolymerization events were observed for the actin cytoskeleton during the salt treatment. Indeed, actin filaments were abundant and maintained to a similar degree as those of the control cells (see mNG-fABD channel in Fig. 7A). Therefore, we conclude that the actin cytoskeleton is more tolerant to salt stress than microtubules.

These results prompted us to investigate whether the actin cytoskeleton has a function in sustaining the microtubule response during salt stress. To assess this further, we looked at both actin cytoskeleton and microtubule dynamics during salt treatment. During microtubule disassembly, small microtubule-related fluorescent foci formed and subsequently began to merge to form larger foci (Fig. 7B). Interestingly, these events coincided with actin filaments (Fig. 7B, Movie 6). Moreover, simultaneous treatment with NaCl and the actin disruptor latrunculin B delayed the formation of larger microtubule-related foci (Fig. 7C). Furthermore, separating microtubule-related foci associated with, and tracked along, actin filaments (Fig. 7D, Movie 7). Again, the disruption of the actin filaments caused a substantial delay in microtubule recovery (Fig. 7E). We therefore propose that actin filaments facilitate rapid re-emergence of microtubules during salt exposure.

Rice morphology determinant (RMD) encodes the rice formin 5 protein, which binds to and bundles actin filaments (Zhang et al., 2011). RMD controls many aspects of rice biological processes, including light responses (Fig. 8A; Song et al., 2019). These results make rmd-1 an ideal genotype to assess the actin array organization and behaviour using our actin fluorescent markers. We therefore crossed the rmd-1 mutant lines with the mNG-fABD-expressing line. We next investigated the actin array in 5-day-old dark- and light-grown rmd-1 coleoptiles. We found that the actin cytoskeletal organization in the rmd-1 mutant was different from that in wild type (Fig. 8B). Detailed inspection of the actin organization revealed less actin bundling in the rmd-1 mutant than in wild type. To quantify any differences in the actin array organization, we again employed Cytoseg2.0 (Nowak et al., 2020). Using this algorithm, we confirmed the reduced actin bundling in the rmd-1 mutant (Fig. 8C). In addition, quantitative analyses revealed that the actin filaments were less coherent in the rmd-1 mutant than in wild type (Fig. 8D) and consequently aligned less well with the major cell axis (Fig. 8E). Interestingly, the actin bundling was significantly reduced in light-grown seedlings compared with dark-grown seedlings (Fig. 8B,C, Table S3), perhaps reflecting a role of actin bundles in rapid cell expansion. Interestingly, this scenario was reversed in the rmd-1 mutant (Fig. 8B,C, Table S3). We also observed the microtubule array in the rmd-1 mutant. Compared with wild type, the rmd-1 mutant displayed some disturbance in the microtubule array, notably in light-germinated coleoptiles (Fig. S7). However, the differences were less pronounced than those observed for the actin cytoskeleton. These differences might explain why the rmd-1 mutant displays a shoot-bending phenotype in light (Fig. 8A).

We established bright and stable fluorescent cytoskeleton markers in rice to facilitate live-cell imaging of the actin and microtubule cytoskeletons (Figs 1, 2). We found that both Ubi1 and RUBQ2 promoters can be used to drive fABD and MAP4 to get homogeneous, bright and stable actin and microtubule fluorescence. For microtubule labelling, we advise using the Ubi1 promoter because the high expression of the RUBQ2 promoter tends to cause abnormal plant phenotypes (Fig. S1). Although the 35S promoter appears to induce silencing in many rice tissues, it generated good fluorescent signals in stomata (Fig. 1) and could be useful for stomata functional studies. In contrast to Arabidopsis, the use of rice tubulin subunits did not label microtubules well in planta (Fig. 1B) and caused phenotypic deviations from wild type. Other proteins, such as Lifeact, mTalin and KMD, etc. (Kost et al., 1998; Vidali et al., 2009; Deeks et al., 2010), can also be tested as potential rice cytoskeleton-labelling strategies.

Although we observed cytoskeletal fluorescence in developing pollen, we did not observe any signals in pollen mother cells (Fig. S2). It is plausible that this is because the cytoskeleton is too diffuse to be observed reliably with our markers. In future studies, it could be worthwhile to test meiosis-specific promoters to drive fABD/MAP4 to study early rice pollen development, because many rice male sterile-related events are determined at the early microspore stages (Wang et al., 2017; Zhang et al., 2019). Currently, cytoskeleton studies in pollen are mainly focused on pollen germination and pollen tube growth (Chang and Huang, 2015; Liu et al., 2018). By contrast, cytoskeletal function and organization are understudied during pollen grain development. Here, we recorded this process and found that both actin filaments and microtubules change from a disordered to ordered organization (Fig. 3). We note that previous studies did not report the highly ordered/radial microtubule structure that we observed (Zee and Ye, 2000). It is worth noting that this study used immunolabelling to visualize the microtubules. It is therefore possible that certain developmental stages were missed or that the fixation and staining process damaged the microtubule structure, as we found it to be sensitive to stresses (Fig. 4G, Fig. S3). Such differences highlight the advantages of live-cell imaging that more accurately capture the ‘natural’ state of the cytoskeleton.

The interplay between the actin and microtubule cytoskeleton has been a longstanding research topic in animal cells (Dogterom and Koenderink, 2019). In plant cells, scientists have speculated that the two cytoskeletal polymers interact to fulfil cellular activities (Petrášek and Schwarzerová, 2009). Indeed, several proteins have the ability to associate with both actin and microtubules (Zhang et al., 2011; Sun et al., 2017). In addition, repolymerization of actin filaments occurred along microtubules in Arabidopsis cells (Sampathkumar et al., 2011). The tight co-alignment of actin bundles and microtubules in maturing rice pollen that we observed (Fig. 4A-E) adds to our understanding of these connections. Here, we found that actin filaments are necessary to maintain radial microtubule distribution and that microtubules are essential for actin bundle formation (Fig. 4F,G). However, the proteins that facilitate these interactions and the biological function of the radial patterns remain to be addressed.

Soil salinization is a severe agricultural problem in crop production (Mukhopadhyay et al., 2021). Microtubules play an important role in the plant salt tolerance (Chun et al., 2021), and the response of the microtubules to salt has been well documented in Arabidopsis (Wang et al., 2007; Endler et al., 2015). We report a similar response of microtubules during salt stress in rice roots, including depolymerization and subsequent reassembly (Fig. 7A, Table S2). However, we also observed that the re-assembly of microtubules appear to be generated via microtubule-related foci (Fig. 7A, Table 1). We speculate that these foci may act as temporary microtubule ‘reserves’ to recover the microtubule array. Consistent with this, the foci appeared to be guided by the actin cytoskeleton (Fig. 7B,D), a notion supported by actin cytoskeleton depolymerization, which caused significant delay in microtubule recovery (Fig. 7C,E). This emphasizes an important function of the actin cytoskeleton during salt stress. The intact actin cytoskeleton may also be important to maintain cell activity, such as trafficking, during salt exposure. To sum up, by using live-cell imaging of actin and microtubule dual-labelled lines, combined with drug treatment, we found that actin filaments support the formation and separation of microtubule-related foci during salt exposure (Fig. 7B-E), and are important for microtubule recovery. Consistently, several actin-associated proteins positively regulate salt tolerance (Zhao et al., 2013; Sengupta et al., 2019; Wang et al., 2021).

Taken together, we generated rice cytoskeleton fluorescent markers and convincingly show that live-cell imaging of the cytoskeletal components in rice can provide new insights into the function of some of the most important polymers in plant cell biology and development.

All plasmids generated for this publication are based on the vector pCAMBIA1301 (GenBank accession number AF234297.1). For the 35S-driven plasmid, the GFP was inserted at the NcoI site and the fABD/MAP4 was inserted between the BglII and BstEII sites. For the Ubi1-driven GFP lines, Ubi1 and GFP were inserted at the EcoRI and KpnI site, respectively; the OsfABD/OsTUA1 was inserted between the HindIII and PmlI sites. For other plasmids, the promoters Ubi1 or RUBQ were inserted at the EcoRI site; the fluorescent protein-coding sequences mNeonGreen or mScarlet were inserted between the KpnI and BamHI sites; and the fABD/MAP4 was inserted between the BamHI and BstEII sites.

The promoters Ubi and RUBQ were amplified from the vectors pTCK303 (Wang et al., 2004) and pRGEB3.2 (Xie et al., 2015). The licenses for using fluorescent proteins mNeonGreen and mScarlet were purchased from Allele Biotechnology and Pharmaceuticals (San Diego, CA, USA) (Shaner et al., 2013; Bindels et al., 2017). The AtfABD and MAP4 were amplified from the Arabidopsis seedlings of 35S::GFP-fABD (Sheahan et al., 2004) and 35S::mCherry-MAP4 (Gutierrez et al., 2009), respectively. The OsfABD (LOC_Os02g48740) and OsTUA1 (LOC_Os03g51600) were amplified from rice cDNA. All amplification primers are included in Table S4.

The plasmids were transformed into rice calli (Oryza sativa Japonica, cultivar 9522) by agroinfiltration. Transgenic plants were obtained through the skilled technology in our laboratory as reported previously (Cao et al., 2020).

To germinate the seeds, glass Petri dishes were used with a layer of filter paper placed at the bottom. No more than 50 seeds were put on the paper, and deionized water was added until the seeds were just covered. The Petri dish was then put into a Percival (E-36L2, Percival Scientific, Perry, IA, USA) growth chamber (condition: 16 h light/8 h dark and 28°C/23°C) for germination. Aluminium foil was used to cover the Petri dish to induce etiolated coleoptiles. After 4 days of germination, the seeds were used for coleoptile imaging. After germinating for 5 days, the seedlings were transferred into liquid 1/2 Murashige and Skoog (MS) medium for further growth in a black box with a black cover. The cover had small holes to hold the seedlings, letting roots grow into the liquid. These roots were used for imaging. To get adult plants, the rice seedlings were grown in the growth chamber for ∼1 month and then transferred to soil. In the summer, the adult rice plants were grown in the field; in the winter, they were grown in a greenhouse until flowering and seed setting.

For normal imaging, the tissues were usually detached from the entire seedling/plant and the epidermis was imaged. Young stem at tillering stage was sliced into 0.5 cm² pieces of 1 mm thickness. The epidermis (∼0.5 cm²) was taken by tweezers from the coleoptile, and the leaf was grown in less than 10 days. Then, 0.5-1 cm length root tips were imaged at the indicated age in each particular experiment. Anthers were removed from spikelets and directly mounted without any process. Pollen was prepared for imaging as described in the Pollen imaging section.

A 3 mm-thick special metal holder slide, with a round hole (diameter 1.5 cm) in the middle, was used to mount the samples. During imaging, the samples were placed in the hole held by a cover slide stuck on one side of the metal slide. A prepared round agar pad was used to cover the samples.

To image pollen, the preparation was slightly modified from those for normal sample imaging. One drop of water (∼20 µl) was kept at the hole place on the cover slide stuck on the bottom side of the metal slide. Anthers from one spikelet were taken into the water and then the anthers were repeatedly squeezed using a tweezer, releasing pollen into the water. The anther residue was removed. Then, another cover slide was stuck on the top of the metal slide without directly touching the pollen samples, making a free space in between and protecting against water evaporation.

For the drug treatment, the water contained 1 µM oryzalin (19044-88-3, Sigma-Aldrich) or latrunculin B (76343-94-7, Sigma-Aldrich). Once the pollen was released into the drug-containing water, the treatment started and the imaging began immediately.

To germinate pollen, the shedding pollen was scattered on the solid pollen germination medium (Liu et al., 2018). To image hydrated pollen, a small piece of medium containing the pollen was directed taken on the prepared slide for imaging after 5 min. To get pollen tubes, the pollen was incubated in the medium for 1 h at 37°C.

For the NaCl treatment, rice seedlings grown in a black box for 4-5 days were used. The root parts of the seedlings were incubated (80 rpm shaking) in liquid 1/2 MS medium containing 250 mM NaCl (CAS:7647-14-5, Sinoreagent, Shanghai, China). After the indicated incubation times, 0.5-1 cm (from the root tip) roots were detached from the seedlings and mounted on slides with the NaCl-containing medium. The epidermises of roots were imaged.

Unless specified otherwise, images in this article were taken with a spinning disk confocal microscope (SpinSR, Olympus, Tokyo, Japan) equipped with a CSU-W1 disk (Yokogawa, Tokyo, Japan). Both 100× (1.50 NA, oil) and 60× (1.35 NA, silicone oil) objectives were used in an Olympus inverted imaging platform IX83 System. Lasers of 488 nm and 561 nm were used to excite the mNeongreen/GFP and mScarlet proteins, and the emission filters were bandpass 525/50 and 617/73, respectively. Unless specified otherwise, z-stacks were acquired with a step size of 0.5 µm. The time intervals of time series are indicated in the figure legends.

All images shown in this paper were preprocessed in Fiji (Schindelin et al., 2012), which usually included contrast adjustment, background subtraction and maximum projection (unless indicated otherwise). Image analysis was done using unprocessed maximum projections of the acquired z-stacks. Colocalization tests were done using the Fiji plugin JACoP (Bolte and Cordelières, 2006), in which Mander’s coefficient and Pearson correlation coefficient analyses were performed. The microtubule-related foci were analysed using the Python package of skimage.feature.blob_log (van der Walt et al., 2014). The microtubule filament number after salt treatment was analysed using a custom-built script developed in Python 3.8.

Quantification of the cytoskeletal organization for both actin and microtubule marker lines was performed using Cytoseg2 (Nowak et al., 2020). The algorithm uses graph theory to describe and create a connected graph of the organization of the filaments. From the connected undirected graph, all edges of the networks were extracted and analysed in MATLAB (MathWorks, Natick, MA, USA) using four parameters: (1) the filament bundling in the network, given by the edge capacity (edge intensity normalized by length); (2) the network alignment with respect to the major cell axis, given by the weighted mean circular angle (Fisher, 1993; Jammalamadaka and SenGupta, 2001; Berens, 2009); (3) the randomness of the network, given by the weighted circular variance (Berens, 2009); and (4) the perpendicularity of the network, given by the weighted median of all smallest angles between each edge and the major cell axis. For all angle parameters (2-4), the edge angles were weighted by the total intensity of each edge.

The extracted data were analysed by circular statistics (see Fig. S8). Because some networks can show a bimodal distribution (4 in Fig. S8), all angles were multiplied by 2, before estimating the mean circular angle and dividing the result by 2 (angle doubling; Fisher, 1993). Estimating the correct mean circular angle from these simulated networks only works for directed networks (3-5 in Fig. S8). Very random networks (1-2 in Fig. S8) look very similar and because the angle distribution is very uniform, the mean circular angle is not a useful parameter for these networks. Therefore, we introduced the perpendicularity parameter. Here, the smallest angle between each edge and the major cell axis is calculated, restricting all angles to 0-90°. Then, the weighted median of these angles is estimated. In order to avoid confusion with the mean circular angle, the resulting weighted median is rescaled from 0-90° to 0-1. For random networks, the perpendicularity will be ≈0.5 (1-2 in Fig. S8); for directed networks with a 45° angle, the perpendicularity is also ≈0.5 (3 in Fig. S8), but for longitudinal oriented networks the value is close to 0 (4 in Fig. S8) and for perpendicular oriented networks it is close to 1 (5 in Fig. S8).

We thank the Core Facility and Technical Service Center, School of Life Science and Biotechnologies [Shanghai Jiao Tong University (SJTU), Shanghai, China] for assistance with confocal microscopy, and Mingjiao Chen (SJTU) and Xiaofei Chen (SJTU) for the rice callus transformation and transgenic plant selection.