Multicolor 3D-dSTORM Reveals Native-State Ultrastructure of Polysaccharides' Network during Plant Cell Wall Assembly.

The plant cell wall, a form of the extracellular matrix, is a complex and dynamic network of polymers mediating a plethora of physiological functions. How polysaccharides assemble into a coherent and heterogeneous matrix remains mostly undefined. Further progress requires improved molecular-level visualization methods that would gain a deeper understanding of the cell wall nanoarchitecture. dSTORM, a type of super-resolution microscopy, permits quantitative nanoimaging of the cell wall. However, due to the lack of single-cell model systems and the requirement of tissue-level imaging, its use in plant science is almost absent. Here we overcome these limitations; we compare two methods to achieve three-dimensional dSTORM and identify optimal photoswitching dyes for tissue-level multicolor nanoscopy. Combining dSTORM with spatial statistics, we reveal and characterize the ultrastructure of three major polysaccharides, callose, mannan, and cellulose, in the plant cell wall precursor and provide evidence for cellulose structural re-organization related to callose content.

Introduction

The plant cell wall plays key roles in environmental interactions, stress responses, cell-to-cell communication, growth, and morphogenesis. In dividing cells, the completion of mitosis, termed cytokinesis, relies on constructing the cell plate, an intracellular precursor to the cell wall separating the two future daughter cells. The nascent plate assembles as a labile, callose-rich lamella that is subsequently reinforced and replaced with other polymers to form a rigid primary cell wall and so marking the completion of cell division (Drakakaki, 2015). Resolving the native cell plate nanoarchitecture will give insights into the early stages of a 3D polymer network organization that precedes the mature primary cell wall. The composition, architecture, and molecular interaction of hydrogels, such as the cell wall and cell plate, define their biophysical and material properties (Abou-Saleh et al., 2018). The mature primary cell wall has to be strong to sustain a high turgor and allow for remodeling during growth (Chebli et al., 2012). How the cell wall acquires and maintains these sophisticated properties constitutes a fundamental question in plant cell biology. Understanding the in muro cell wall assembly will help elucidate this enigma and inspire the next generation of tissue engineering and smart biomimetic materials (Yao et al., 2020). Despite its fundamental importance, the tissue-context native-state three-dimensional organization of the cell wall/cell plate polysaccharides' network remains largely unknown. Established microscopies cannot resolve cell wall polymers in three dimensions without losing molecular specificity or resolution (Chebli et al., 2012). Nowadays, single-molecule localization microscopies (SMLM) permit experimental access in the realm of molecular assemblies in three-dimensions while preserving biochemical specificity. SMLM finally enables observing native-state polymeric and molecular ultrastructure with nanometer precision. Due to technical limitations (tissue-level imaging, strong autofluorescence), its application to plant science is sporadic (Liesche et al., 2013; Eggert et al., 2014; Dong et al., 2015; Haas, et al., 2020b; Haas, et al., 2020a); however, its recent application is already changing the vision of the cell wall structure (Haas, et al., 2020a). Moreover, the biologically relevant three-color 3D nanoscopy implementation, even in simpler isolated cellular systems, is rare.

Here we harness the full potential of SMLM performing three-color 3D dSTORM (direct stochastic optical reconstruction microscopy) on plant tissue. To do so, we identify the optimal combination of photoswitching dyes for triple-color dSTORM in OxEA (Oxyrase-based) (Nahidiazar et al., 2016) buffer compatible with strongly autofluorescent samples. We compare two alternative 3D-dSTORM configurations (1) astigmatic and (2) biplane imaging and evaluate their performance in thin tissue cuts. Using dSTORM nanoscopy we visualize the assembly of three major polysaccharides, cellulose, mannan, and callose, in the nascent cell wall. By employing spatial statistics, we characterize the incorporation, arrangement, proximity, and turnover of these polysaccharides during cell plate development and show their structural transition during plate maturation.

Results

Optimizing the Imaging Conditions for Tissular Multicolor 3D dSTORM

The resolution of the reconstructed SMLM images depends on (1) the labeling density and (2) the localization uncertainties determined by the photon noise, pixel noise, and background noise (Thompson et al., 2002; Mortensen et al., 2010). The dye photoswitching properties, including its brightness, on-off cycle time, and survival time, are critical for gaining high spatial resolution (Dempsey et al., 2011). Therefore, the buffer composition and optimal dye combination for multicolor 3D-dSTORM in biologically relevant applications are essential to maximizing the resolution (Heilemann et al., 2008; Dempsey et al., 2011; Van De Linde et al., 2011; Lehmann et al., 2016; Nahidiazar et al., 2016). Previously we showed a two-color 3D-dSTORM on plant tissue sections using 2F4-GLOX (glucose oxidase-based, 2F4 is a T/Ca/S buffer, see Transparent Methods) buffer (Haas, et al., 2020a, 2020b). Here we tested the OxEA buffer and defined the optimal conditions for three-color 3D-dSTORM observations. We choose Alexa 647 for far-red detection, performing well in both OxEA and GLOX (Lehmann et al., 2016; Nahidiazar et al., 2016; Haas et al., 2018; Haas, et al., 2020a); for red detection, we choose CF568, which displayed superior photoswitching in our modified OxEA buffered with 1 × 2F4 solution. Next, we identified the optimal green dyes, particularly challenging for plant tissue due to strong autofluorescence. Figure 1 presents three-color 3D-dSTORM astigmatic imaging of cell plates from the shoot apical meristem and organ primordia in Arabidopsis thaliana. Figure 1A compares a standard 2D pixelated dSTORM image with a coordinate-based scatterplot allowing for three-dimensional “pointillist” data representation and opening a broad avenue for statistical quantifications (Figure S1). To compare photoswitching properties, we imaged four different cell wall epitopes, namely crystalline (CBM3) and amorphous (CBM4) cellulose, hetero-mannans (PDM), and callose after the enzymatic pectin extraction (Haas, et al., 2020b), and tagged them with different fluorescence dyes (Handford et al., 2003; Hernandez-Gomez et al., 2015). Tested dyes, except Alexa 488, displayed very good photoswitching in OxEA at pH 8.5 buffered with a 2F4 solution. This allowed a high-quality super-resolution reconstruction visible in the plots of developing plates characterized by a foam-like structure with voids excluding any of the studied epitope (Figures 1A, 1B, and S2). Green fluorescence dyes such as fluorescein isothiocyanate (FITC), Chromeo 505 (CH505), and CF488 in OxEA, despite the higher autofluorescence in this spectral region (Video S1. Extracted image sequence showing CBM3 tagged with FITC and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1, Video S2. Extracted image sequence showing Callose tagged with CF488 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1.d, Video S3. Extracted image sequence showing Callose tagged with Chromeo505 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1, Video S4. Extracted image sequence showing PDM tagged with Alexa488 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1, Video S5. Extracted image sequence showing Callose tagged with ATTO488 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1, Video S6. Extracted image sequence showing CBM3 tagged with Alexa647 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1, Video S7. Extracted image sequence showing CBM4 tagged with Alexa568 and acquired at ∼60 fps with an astigmatic approach. Related to Figure 1.), perform as well as either Alexa 647 (A647) or CF568, whereas in the most frequently used GLOX buffer they display no or poor photoswitching (Dempsey et al., 2011). We achieved <10 nm accuracy when determining a fluorophore's position (localization precision) for different dyes and cell wall targets separately, resulting in a resolution of ∼40 nm in xy and ∼70 nm in z (Figures 1C and S3).

Figure 1

Optimizing Dye Combination for Three-Color 3D-dSTORM

(A) DSTORM data representation as a pixelated 2D image and coordinate-based 3D scatterplots.

(B) Representative cell plates with different epitope-dye combinations showing amorphous cellulose (CBM4, violet), mannan (PDM, orange), and callose (green).

(C) The localization precision distributions for different epitopes and photoswitching dyes. The distributions include localizations with at least 1,500 photons detected.

Comparison between Astigmatic and Biplane 3D dSTORM Modalities

Next, we compared the performance of two 3D dSTORM methods, astigmatic and biplane imaging (Huang et al., 2008; Juette et al., 2008). We combined astigmatism with oblique illumination, which confines the imaging section, lowering the out-of-focus signal. In contrast, the biplane imaging uses epifluorescence mode, which can increase the background signal, especially for strongly autofluorescent samples, but allows for deeper observations. Figure 2 shows distinct cell plates for a biplane (A) and astigmatic (B) imaging of callose detected with A647 or CF568 (light-to-dark-blue colormap encodes the z-coordinates), cellulose (CBM3) detected with CH505 or CF488 (orange), or mannan (PDM) detected with CF568. Figure 2C shows the distribution of the axial coordinates of localized fluorophores for both modalities. The z-detection range for astigmatism was <800 nm; in comparison, biplane illumination displayed a ∼50% wider z-detection range. The broader and more homogeneous z-detection range is the main advantage of the biplane imaging mode over the astigmatism-based approach. However, astigmatism-based imaging is simpler and more versatile, permitting variable illumination modes: total internal reflection fluorescence microscopy and oblique and epifluorescence illumination. Finally, in our conditions, the biplane approach offered a slight improvement in the axial localization. To achieve a similar axial resolution with astigmatism, we filtered the localization with >2,500 photons (the localization precision scales with the inverse square root of the number of photons) (Figures 2B and S7).

Figure 2

Comparison of Astigmatic and Biplane 3D dSTORM Imaging Modalities

(A–C) Representative cell plates imaged with (A) biplane and (B) astigmatic mode. Orange marks cellulose (CBM3) in (A) and mannan (PDM) in (B). Callose is represented with light-to-dark blue colormap encoding z-position. Data in (A) represent all detected localizations and in (B) localizations with at least 2,500 photons detected. In (A) and (B), the inset images show the conventional oblique illumination pictures of the area imaged with dSTORM. (C) Bean plot of the axial detection range for the astigmatic and the biplane imaging for different photoswitching fluorophores showing all unfiltered localized molecules. The black arrow in (A) indicates a callose-enriched rim protruding from the primary cell wall and directly linked to the developing plate. Scale bars, 1 μm in scatterplot and 5 μm in image.

Visualizing Cell Plate Assembly with Three-Color 3D dSTORM

The different cell plate developmental stages, from the coalescence of the Golgi-derived vesicles and the formation of a membranous tubular network to the maturation stage, have been observed using electron and fluorescence microscopy techniques (Frey-Wyssling et al., 1964; Seguí-Simarro et al., 2004; Baluška et al., 2005; Thiele et al., 2009; Miart et al., 2014; van Oostende-Triplet et al., 2017). However, these techniques lack either molecular specificity or the required resolution to observe the polymers' network in three dimensions. Multicolor 3D-dSTORM allowed us to visualize the polymers' nanoarchitecture and their mutual spatial arrangement in 3D space. Their accumulation level and spatial distribution in dividing cells can serve to discriminate different cell plate developmental stages. First stage is the gradual insertion of polysaccharides, such as callose deposition into small patches (Figure 3, plates 1–4, and Figure S4), that subsequently extend and peak at the fenestrated planar sheet stage (Figure 3, plates 5–6, Figure S5). Accurate determination of the width of this callose-rich sinuous lamella (>∼40 nm, Figure S3D) requires knowing its orientation in three dimensions; otherwise, the projection errors would lead to >1 μm width measurements (Figure S1B).

Figure 3

Visualizing Cell Plate Assembly with Three-Color 3D dSTORM

Representative cell plates showing mannan (PDM-CF568, orange), callose-A647 (green), and cellulose (CBM3-FITC, violet). Top row, conventional oblique illumination images with numbers marking cell plates presented below as dSTORM scatterplots shown in two views: top view 2D (xy, the same as low-resolution images above), and side-on 3D view, where the z axis is oriented perpendicular to the conventional image plane. The red arrowheads mark granular structures in close vicinity to the forming plates that may represent mannan-loaded vesicles. The blue arrows show newly synthesized callose and cellulose at the membrane surrounding the forming plate, but could also represent synthesis inside vesicles, as evidenced previously (Chebli et al., 2012). For cell wall #8, double arrow marks vertically aligned aggregates of STORM signal points, corresponding to CBM3 fibers. A black arrow marks callose at a pit. Scale bars, 1 μm for top view plots and 5 μm for low-resolution images.

In Muro Biochemical Analysis Using the Point Correlation Function

Previous root tip and tobacco cell culture studies showed that the cell plate grows centrifugally, i.e., commencing from the cell center and then outward until meeting with the walls of the parental cell (Samuels et al., 1995; van Oostende-Triplet et al., 2017). Interestingly, we observed that, in the apical meristem and organ primordia, cell plates often protrude from opposite paternal cell walls and accrete at the cell center or grow from just one edge (Figure 3; plates 1, 2, 5, 6, and Figures S4 and S5) (Yang et al., 2016). Next, we estimated statistical dependence between the accumulation of different polymers in the developing plates using pairwise linear regression between the number of detected epitopes. This analysis showed that the accumulation of studied polysaccharides (CBM3/4, PDM, callose) is positively correlated, concomitant, and not sequential (Figure 4A). Deviation from the linear dependence, especially between callose and cellulose (CBM3 staining), reflects the cell plate maturation stage, going from a “fluid” and wrinkled lamella to a stiff, flat primary cell wall. Callose is digested away during maturation, but cellulose accumulation keeps increasing or levels off (Figure 3; plate, 7).

Figure 4

In Muro Biochemical Analysis Using the Point Correlation Function

(A) Pairwise linear regression analysis for the relative number of epitopes performed between callose-A647, mannan (PDM-CF568), and crystalline cellulose (CBM3-FITC) for the left panel and the amorphous cellulose (CBM4-FITC) for the right panel; each point represents a cell plate. The legend indicates the Spearman correlation coefficient between pairs of epitopes.

(B) Three different classes of bivariate point patterns and corresponding graphs of PCF. From the left: independent (PCF = 1 at all distances), segregated (PCF <1), and aggregated (PCF >1).

(C) Bivariate pair correlation functions (PCF) for the selected plates shown in Figure 3. The dashed black line marks experimental PCF; the solid blue line and the shaded blue area mark the average PCF and 95% confidence intervals over 25 randomization steps, respectively. The red line marks the fit to exponentially modified Gaussian function; d12 is a correlation distance recovered from the fit.

(D) Univariate pair correlation functions (ACF) for the selected plates shown in Figure 3. The dashed black line marks experimental ACF; the solid blue line and the shaded blue area mark the average PCF and 95% confidence intervals over 25 randomization steps, respectively. The red line marks the fit to the sum of Gaussian function and exponentially modified Gaussian function; d is the true domain size and N is the number of molecules per domain recovered from the fit.

The “pointillist” dSTORM data permit the quantification using spatial point pattern statistical descriptors to characterize the spatial arrangement of different epitopes (Sengupta et al., 2011; Shivanandan et al., 2016). We developed a boundary-corrected 3D bivariate pair-correlation function (PCF) analysis, commonly used in cosmology. The PCF-based analysis is robust to estimate spatial arrangement and domain size without setting arbitrary thresholds (Sengupta et al., 2011; Veatch et al., 2012; Sengupta et al., 2013). The 3D PCF estimates the number of type 1 molecules within a thin spherical shell centered at the type 2 molecule, normalized to the expected number under the independent distribution. The empirical PCF was compared with the randomized PCF, calculated by the repetitive random reallocation of the experimental points within the 3D boundary enclosing the plate (Figure S8). Randomized distributions have the average PCF = 1, indicating spatial independence between two epitopes (Figure 4B). If the experimental PCF lies above the 95% confidence interval (blue shaded area) at a given distance, it suggests a spatial association. Figure 4C shows that callose with mannan (PDM) and callose with cellulose (CBM3) remain correlated at all developmental stages. At early stages, PDM and CBM3 show nearly independent distribution (Figure 4, plates 3 and 4), which increases during the callose-rich phase, to become again almost independent at the mature plate and primary cell wall stage. Similarly, the correlation between callose and mannan increases from an early stage, peaks at the callose-rich stage, decreases at a mature plate, and becomes nearly independent in the primary cell wall. When applied to the univariate (single molecular species) point patterns, PCF is an autocorrelation function (ACF), which can yield information on the characteristic scales of the molecular assemblies. The effective domain size or the correlation length r0 can be estimated using a double exponential fit to empirical ACF (Equation 3, Transparent Methods); however, it does not account for single-molecule over-counting, and thus it does not permit stoichiometry evaluation (Figures S8F, S8G, S9, and S10). Due to the natural variability of bound fluorophores, the finite localization precision, and fluorophore blinking, the individual fluorophore appears as a cluster of multiple points (Figures S3E–S3H). Previous reports proposed a model that accounts directly for the fluorophore blinking and the finite localization precision (Sengupta et al., 2011; Veatch et al., 2012; Sengupta et al., 2013), bringing single-molecule imaging to stoichiometric quantification. This model includes the contribution from the multiple detections of a single dye and the target molecule organization, which are convolved with the ACF of the point spread function (see Equation 4, Transparent Methods). This approach estimates the true domain size, d, and the average number of molecules, N, present within that domain. This model was developed for PALM, and it does not account for the multiple tagging of the secondary antibodies. When we performed our experiments, the secondary antibodies conjugated to exactly one dye molecule did not exist for all species used in this study. Previous dSTORM applications bypassed this issue by generating control constructs using SNAP and Halo tags that assures precise tag-to-target stoichiometry (Zhao et al., 2014). Unfortunately, polysaccharides cannot be tagged by genetic means. We attempted to estimate the number of conjugated dyes with the spatial clustering using Voronoi diagrams and Delaunay triangulation followed by temporal clustering (Zhao et al., 2014) using kernel density estimates applied to data generated by the isolated secondary antibodies (Figure S3). This analysis shows that the small clusters contain localizations assigned to one temporal cluster (one burst), and the larger clusters contain few temporally separated bursts. The average number of bursts per antibody was 2 for Alexa 647- and CF568-tagged antibodies and 1 for FITC-tagged antibody, which could serve as a rough estimate of the number of detected dyes.

The effective domain size r0, the true domain size d, and the number of molecules per cluster, N, that we estimated for mannans was r0 ∼28 nm, d ∼16 nm, N = 7; r0 ∼30 nm, d ∼29 nm, N = 14; and r0 ∼24 nm, d ∼25 nm, N = 5 at the early, mature, and primary cell wall stage, respectively; at the callose-rich stage it was r0 ∼70 nm, d ∼78 nm, N = 15–77 (Figures 4D and S8; see Transparent Methods for details). This reflects mannans' gradual accumulation and its subsequent removal. However, masking of mannans by other carbohydrates cannot be excluded. Importantly, r0 ∼24 nm corresponds to the value obtained for the single isolated secondary antibody (Figure S8F) and can serve as an estimate for the resolution limited by the localization precision, but not the linkage errors (the size of antibodies' complex). This is supported by the observation that r0 and d are similar for the values >20 nm.

Similar dependencies are observed for callose and cellulose. We observed r0 ∼66 nm/66 nm, d ∼47 nm/50 nm; r0 ∼114 nm/187 nm, d ∼97 nm/171 nm; r0 ∼64 nm/169 nm, d ∼46 nm/149 nm; and r0 ∼48 nm/94 nm, d ∼41 nm/67 at the early, callose-rich, mature plate, and primary cell wall stages, respectively, for callose/cellulose. At the primary cell wall stage, the remaining callose patches locate to plasmodesmata that comprise the symplastic connections between the two future daughter cells (Figure 3; cell wall 8, Figure S6). Interestingly, the correlation length of cellulose also decreases between the mature plate and the primary cell wall. This suggests that the cellulose reorganizes from uniform to more fibrous upon callose digestion. These observations agree with in vitro studies showing that the addition of callose to cellulose reduces the hydrogel elastic modulus and disrupts the cellulose network (Abou-Saleh et al., 2018). It is further supported by the nanoimaging of callose deposition and its association with cellulose during fungal invasion (Eggert et al., 2014).

Contrary to ACF, PCF is not dependent on the multiple counting of the same dye molecule (Sengupta et al., 2013) and can be modeled by the convolution of exponential decay and the cross-correlation of the point spread functions of each fluorophore (see Equation 6, Transparent Methods). This model permits recovery of the length scale d12 at which two molecules are correlated (Figure 4C).

Discussions

We demonstrate that multicolor 3D-dSTORM, coupled with statistical quantification, is a powerful bioanalytical tool to study the ultrastructure of cell wall polysaccharides at the tissue level and with molecular specificity. The nanoimaging in three dimensions is indispensable for an accurate reconstruction of the cell wall molecular assemblies and assessing their spatial relationship. 3D-dSTORM permitted cellulose re-organization observations during cell plate maturation, which could explain the mechanical changes between the callose-rich flexible plate and stiff primary cell wall. In the apical meristem, we showed that cell plates often grow out from a “rim” connected to the paternal cell wall, not from the center of the cell as in the root meristem. Moreover, we show that callose, cellulose, and mannan accumulation are concomitant but occur at different rates and results in different final densities. We observed that mannan localizes to small clusters, with the density peaking at the callose-rich stage. The callose accumulation plateaued at the callose-rich stage, whereas the cellulose leveled off at the mature stage. To facilitate the absorption of SMLM into plant science, we optimized buffer/dye combination for three-color imaging and presented two 3D-dSTORM methods, astigmatic and biplane imaging. Due to the distance between epitope and signal (the length of the probe plus antibodies), our resolution (xy∼40 nm, z ∼70 nm) is not localization precision-limited (∼10–20 nm). The development of nanobodies, small tags, and oligosaccharide probes (Mravec et al., 2014) will permit routine dSTORM imaging with single-digit nanometer precision. SMLM technology continuously improves; recently, a new technique, MINFLUX, has demonstrated 1–3 nm localization precision (Gwosch et al., 2020). The combination of dSTORM and expansion microscopy (ex-dSTORM) can further push molecular resolution (Gambarotto et al., 2019). It is reasonable to expect that future experimental advances will bring localization nanoscopy to a genuinely molecular level.

Limitations of the Study

Multicolor 3D dSTORM allows fine mapping of the cell wall polymers' intricate ultrastructure at nanometer-range precision and with biochemical specificity. However, this approach can only map the static, steady-state structure. Moreover, it relies on immunolabeling using full-length primary antibodies conjugated with secondary antibodies that together introduce 15–25 nm linkage errors. The development of smaller nanoprobes that target cell wall polymers is essential to bring the resolution to a near-molecular level. Furthermore, the application of secondary nanobodies with the precise number of conjugated fluorophores will enable truly stoichiometric imaging. In addition, the 3D Point Pattern Analysis performed in this study did not take into account the anisotropic structure of the studied polysaccharides. Finally, cellulose labeling necessitated enzymatic extraction of pectin that may influence the structure of the studied polysaccharides. In addition, other polysaccharides may still mask some of the target epitopes that were observed in this study.

Resource Availability

Lead Contact

Kalina Tamara Haas kalina.haas@inrae.fr.

Material Availability

This study did not generate any new material.

Data and Code Availability

All software used in this study are available at the GitHub repository https://github.com/inatamara/Grafeo-dSTORM-analysis-. All the dSTORM images will be provided upon request.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.