Visible features of the plant, such as flowers, leaves, and roots, result from microscopic processes of tissue and organ formation, which are in turn driven by cellular and molecular dynamics. How whole plants and their organs, tissues, cells, organelles, and sub-cellular molecules interact within three-dimensional (3D) space is crucial to understanding plant biology across diverse genetic backgrounds and changing environments. Capturing 3D external and internal volume data brings an important spatial component to plant phenotype analysis that can help identify genetic factors that control trait development not manifested in simpler 2D images. Recently, 3D imaging platforms and corresponding phenotyping methods have developed rapidly, and many approaches have been applied to a wide diversity of plant growth and development phenomena (Morris et al., 2017; Perez-Sanz et al., 2017; Tracy et al., 2017; Jiang et al., 2019; Roth et al., 2019; Strock et al., 2019; Czymmek et al., 2020; Li et al., 2020). Furthermore, sophisticated analysis pipelines have been developed to extract and quantify data from relevant features, with improved imaging leading directly to improved trait analysis and models (Bassel and Smith, 2016; Koebernick et al., 2017; Mathers et al., 2018; Jackson et al., 2019; McKay Fletcher et al., 2020; Sultan et al., 2020; Théroux-Rancourt et al., 2020; Wolny et al., 2020).

Current 3D systems use photons, electrons, and X-rays for imaging biological samples. Photon-based tomography (confocal, multiphoton, light-sheet, super-resolution) generates valuable 3D information with chemical specificity based on fluorescent probes and/or auto-fluorescence in both living and fixed samples (Wymer et al., 1999; Feijó and Moreno, 2004; Schubert, 2017; Ovečka et al., 2018). However, imaging with photons fundamentally has sample size constraints (Prunet and Duncan, 2020; Rawson et al., 2020) and nondestructive approaches (but see (Strock et al., 2019) are typically limited to shallow image depths for plant tissues as they are highly refractive due to cell walls, vacuoles, and air spaces, that often require special preparation or clearing to mitigate (Littlejohn et al., 2010; Kurihara et al., 2015). Another approach, 3D volume scanning electron microscopy (SEM), repeatedly images a resin block-face after removing thin layers via an in situ microtome or surface ablation via focused ion beam (Bhawana and Cahoon, 2014; Kittelmann et al., 2016; Czymmek et al., 2020; Harwood et al., 2020). With volume SEM, detailed 3D spatial information is possible down to a few nanometers; however, practical imaging volumes are typically restricted to a maximum of a few hundred micrometers (Kubota et al., 2018). Additionally, sample preparation is highly specialized, where prolonged protocols with greatly increased sample staining are required to ensure adequate signal-to-noise and sample resolution.

The X-ray microscope (XRM) is a specialized X-ray computed tomography (CT) instrument that incorporates microscope objective lenses (Jacobsen, 2020). In XRM, 2D digital radiographs are projected onto a scintillator-coated objective lens where the X-ray signal is converted into visible light, magnified, and the resulting image collected with a detector, typically a CCD camera. The objective lens in the beam path increases the resolution of the XRM compared with conventional X-ray CT systems, which rely solely on geometric magnification, and multiple objective lenses make possible the collection of 3D volumes over a wide range of sample sizes and resolutions not practical with other imaging platforms. Furthermore, in contrast to photon- and electron-based tomography, where imaging penetration depths are restricted to one or a few cell layers, the ability of X-rays to penetrate through large, intact and highly light scattering plant structures is a distinct advantage. Thus, lab-based XRM provides a critical link between photon and electron 3D imaging, establishing high-resolution intermediate magnification levels for whole plant imaging.

X-ray imaging contrast is based on the differential density of features within a specimen; however, most biological samples are composed of relatively uniform low-density organic material, often with insufficient contrast to generate useful 3D image data. For this reason, sample fixation protocols and contrast agents play important roles in imaging biological samples with XRM (Metscher, 2009; Rawson et al., 2020). Staedler et al. (2013) evaluated conventional EM sample fixation methods and contrast agents as tools for examining plant samples with XRM. These methods were sufficient for the relatively short scan times necessary to capture floral anatomy for analysis using basic morphological markers, but do not provide adequate stability for the longer scan times required for high magnification/resolution of more subtle and complicated plant features.

Here, we have built upon these earlier methods, incorporated improvements in sample preparation, and leveraged instrument technological advances to expand the range of plant cells, tissues, and organs that can be imaged at high resolution with XRM. This enabled unprecedented 3D perspectives and insights for phenotype and form-function studies. Further, due to the improved image quality of 3D volume data, advanced image analysis and cell-level segmentation of relevant structures of interest using computer vision and learning are substantially enhanced. We applied XRM with improved sample preparation for multimodal and multiscale 3D imaging, where samples prepared for high-resolution volume SEM were first imaged with XRM as part of a correlative workflow (Bushong et al., 2015; Tsang et al., 2018; Mitchell et al., 2019; Bayguinov et al., 2020). Taken together, we anticipate that the adoption of these improved techniques will make an important contribution to plant biology, expanding the reach of XRM as a routine tool for 3D imaging for plant scientists.

Rapid technological and computational improvements over the past several decades have enabled the acquisition and analysis of 3D volumetric data of ever-increasing size and complexity for plant biology. However, intact plant structures are often difficult to image in 3D with light and/or electron microscopy platforms due to challenges including relatively large sample size, an optically scattering and chemically resistant cuticle, cell walls and air pockets that impede fixation and exogenous treatments, complex morphologies, and delicate features. As a result, a notable gap exists with light and electron imaging modalities that restrict high-resolution tomographic imaging to small samples that provide limited contextual information relative to the whole organ or plant.

Because of its ability to penetrate dense tissues, conventional X-ray CT was utilized to address some of these imaging limitations in plant science as early as the 1990s (Aylmore, 1993; Heeraman et al., 1997) and 2000s (Stuppy et al., 2003; Kaminuma et al., 2008; Leroux et al., 2009; Dhondt et al., 2010; van der Niet et al., 2010). However, most conventional X-ray CT instruments rely solely on geometric magnification, and thus have functional resolution and sample size limits based on source-sample-detector geometry. Synchrotron-based X-ray CT has advantages in shorter scan times, high-resolution imaging, and allows element mapping using X-ray fluorescence (Kopittke et al., 2018); however, their limited availability relative to lab-based XRM restricts widespread utilization in plant biology. Here, we demonstrated that lab-based XRM could image intact plant structures in fine detail, and subsequently zoom to selected regions of interest down to the cellular and subcellular level, effectively bridging the tomographic imaging gap of light and electron microscopy.

Contrast and stabilization of samples are critical for high-resolution X-ray imaging due to the longer scan times. Previous XRM studies used conventional sample fixation, contrast enhancement, dehydration, and critical point drying followed in some cases by sputter coating (Staedler et al., 2013; Bellaire et al., 2014; Jeiter et al., 2018; Jeiter et al., 2020; Vasile et al., 2021). Other researchers stabilized samples in the contrast enhancement medium, sealed within pipette tips or polypropylene tubes (Gamisch et al., 2013; Staedler et al., 2013, 2018; Reich et al., 2020). In these studies, the goal was to survey a relatively large number of samples and evaluate basic 3D shapes using scans no longer than necessary to achieve stated research aims in a timely manner. With plant tissue prepared in this way, even with floral samples stabilized in polyester fibers or in ethanol, sample drift during longer XRM scans can be problematic and therefore limit achievable resolution and magnification. To address this challenge, we specifically developed a simple but effective strategy whereby fixed samples were contrasted in ePTA and stabilized in LMP agarose inside the smallest possible tube relative to sample size. This maintained a vertical orientation of plant samples for optimal rotational axis during imaging. LMP agarose effectively coated and stabilized even the most delicate or complicated samples, and provided complete physical support preventing tissue movement and drift during long scans. Higher scan magnifications and resolutions required longer scan times and our methods allowed both, which concomitantly enabled improved computational segmentation by providing higher quality raw image data. One recent example has been reported where conventional X-ray CT was used to image fresh plant tissue using molten petrolatum to stabilize small aquatic plants (Jones et al., 2020), although without contrast enhancement of samples it remains unclear if this method would be suitable for high-resolution XRM imaging.

Contrast enhancement with PTA has widespread utility due to broad staining of various plant tissues (Hayat, 1993; Staedler et al., 2013). As effective as PTA is as a contrast agent for most plant samples, it requires a lengthy incubation period—at least 14 d in our methods—to generate sufficient contrast. The application of microwave tissue processing could potentially accelerate PTA contrast enhancement protocols down to days or even hours (Kang et al., 1991; Zechmann and Zellnig, 2009; Carpentier et al., 2012). In our hands, conventional fixation in buffered aldehydes followed by post-fixation with osmium tetroxide and potassium ferrocyanide was effective in contrasting samples like root tips and Arabidopsis plantlets, but penetration was limited into larger complex floral structures. Iodine-based contrast agents can rapidly contrast entire plant tissues within 24 h, but they easily leach out of specimens into the surrounding agarose, lowering effective sample contrast and increasing scan noise and background. Alternatively, more rigid plant specimens can be contrasted in iodine and instead of agarose, stabilized with relatively X-ray transparent polyester fibers or small polystyrene beads, adding pieces of paper towel or lab wipes saturated with water or ethanol to maintain high relative humidity before sealing in sample tubes. Other robust plant materials such as wood samples (Reynolds et al., 2018) or seeds and grains (Su and Xiao, 2020; Han et al., 2021; Matsunaga and Smith, 2021) are readily imaged with XRM with a minimum of preparation. Supplemental Figure S8 shows XRM scans of maize, sorghum, soybean, and wheat seeds where no fixation or contrast enhancement were used. Seeds were simply placed in PCR or Eppendorf tubes and stabilized with expanded polystyrene beads, to observe a range of internal seed structures.

Meristematic structures, which are often too large and dense for meaningful 3D imaging using fluorescence or EM, can greatly benefit from XRM. Numerous recent studies have provided important insights into meristematic and inflorescence structure development, including S. viridis (Yang et al., 2018), Hordeum vulgare (Walla et al., 2020), maize (Teng et al., 2020), Aquilegia coerulea (Min and Kramer, 2020), and other grasses (Kellogg et al., 2013; AuBuchon-Elder et al., 2020). These examples illustrate the breadth of research questions that can be enhanced by additional tissue- and cell-level 3D XRM imaging to measure and evaluate meristematic structure development over scales not possible or practical with other methodologies. To address the challenge of meristematic cells having a relatively uniform density between the cytoplasm and cell walls, nuclei can be used as foci for segmentation since ePTA imparts high contrast to nucleic acids and the location of the nucleus in the cell can assist computational efforts to identify individual cells. This may require higher magnification/resolution scans to improve segmentation. Segmentation of meristematic tissues nonetheless remains challenging even with high-resolution scans of well-contrasted samples, each new sample requiring a unique combination of automated and manual segmentation solutions to accurately and reliably separate and identify specific cells and tissues. Additional computational tools for cell segmentation are emerging and should provide powerful options for the analysis of 3D volumes from a wide range of plant samples (Wolny et al., 2020).

The use of contrast agents for meristematic tissue is critical if cell-level segmentation is the goal. Thus, the development and application of cell wall-specific chemistries such as electron-dense probes or particles to generate XRM contrast agents that are specifically selective for plant and fungal cell walls would be highly beneficial. While selective staining may be possible using affinity-based small molecules (e.g. gold nanoparticles), we also envision the application of genetically encoded expression of ascorbate peroxidase-fluorescent protein fusions to a library of select plant cellular compartments (Ariotti et al., 2018). The ascorbate peroxidase method works by producing localized expression of an osmiophillic precipitate that can be contrasted by light, X-ray and electron microscopy when the ascorbate peroxidase fusion protein is incubated with 3,3-diaminobenzidine (DAB) and hydrogen peroxide (Martell et al., 2017). This would allow osmium-reactive DAB deposits at target structures that could be visualized first by XRM and subsequently processed for high-resolution EM studies as described previously (Ng et al., 2016; Tsang et al., 2018; Kong et al., 2020).

Plant–microbe interactions, root biology, and vascular bundles were readily differentiated by XRM imaging with relatively straightforward sample preparation. For example, root nodules induced on soybean were visualized in detail (Figure 4, A–D) without the time consuming and technically challenging tasks of physical embedding and sectioning, acquiring 2D photomicrographs of each section, and manually assembling serially organized images into a 3D model to study root nodule morphology and vasculature (Livingston et al., 2019). Mycorrhizal fungi that form arbuscules within cortical cells of plant roots for mutualistic nutrient exchange between plant and fungus can be visualized and quantified as shown by segmentation of R. irregularis in maize roots (Figure 4, E and F). We also developed a syringe barrel system where both host root system architecture and fungal symbiont were imaged in situ (Figure 4, G and H; Supplemental Figure S7) without perturbation of the local environment. We are currently exploring various physical and chemical affinity probes and contrast agents that could be applied directly to the growth medium for improved XRM visualization of root–microbe interactions, adapted from Scotson et al. (2019) who used X-ray CT and SEM to evaluate gold nanoparticles as a contrast agent in soil, studying the interaction of particles from a liquid suspension drawn through a column containing various media.

Cell-level measurement of roots in 3D is now possible with XRM imaging to better understand how simple root growth responses such as elongation, bending, and branching result in complicated and vast root networks that occupy square meters of soil (Figure 3, A–D; Supplemental Figures S5 and S6). This enables studies that compare cell number, morphology, and 3D position in root tips between maize lines known to differ in root system architecture at relevant developmental stages (Jiang et al., 2019), or for modeling hydraulic conductivity (Heymans et al., 2020). Iodine-based contrast agents were useful for visualization of vascular elements (Figure 3E), and coupled with XRM will be of great value for mapping physically complex regions such as the root–shoot transition zone (nodal plexus) and graft junctions between rootstock and scion. We are currently involved in multiple collaborations using XRM to image and evaluate specific genetic effects on morphology and development of samples across a range of plant species, which we anticipate will further demonstrate specific uses of XRM for generating biologically meaningful insights.

Imaging of EM prepared resin-embedded specimens in the life sciences increasingly has been applied to 3D XRM (Hanssen et al., 2012). Likewise, protocol improvements in SBFEM sample preparation techniques (Wanner et al., 2015; Titze and Genoud, 2016) including the protocol used here with en bloc metal staining (Hua et al., 2015), demonstrated improved osmium penetration and enhanced contrast in thick plant tissues which inherently benefit high-resolution XRM and SBFEM correlative workflows (Bushong et al., 2015). While there have been a handful of published plant-based SBFEM examples (Kittelmann et al., 2016; Oi et al., 2017; Czymmek et al., 2020; Harwood et al., 2020), we reasoned that further improvements in uniformity, conductivity and intensity of plant staining (Deerinck et al., 2018) would substantially benefit both XRM and SBFEM image contrast and quality for correlative studies. Additionally, specimens encapsulated in rigid hard formulation epoxy resins provided excellent stability for long-duration XRM scans, translating to the optimal resolution. Our protocol resulted in optically opaque specimens in resin blocks that initially were scanned with XRM to identify regions of interest and guided our subsequent high-resolution volume EM, all from the same block and using a single optimized sample preparation method. Notably, we believe these data reflect the highest resolution reconstruction of subcellular plant structures via microCT or XRM published to date, allowing sub-organellar starch segmentation via deep learning (Figure 5G). We envision that our plant-based correlative XRM to EM workflow will serve as an effective way to obtain 3D contextual histology, as a standalone approach, or modified to include other correlative strategies such as targeting genetically encoded structures and/or combined with light microscopy approaches. While less accessible, we acknowledge that further resolution gains using synchrotron-based platforms will allow nanoscale imaging (Hwu et al., 2017; Fonseca et al., 2018; Chin et al., 2020) also benefiting from ours and others’ plant sample preparation strategies. Ultimately, these tools and protocols will accelerate the ease in localizing, resolving, and quantifying important discrete and multiscale plant biological phenomena, including developmental studies, phenotype characterization, and plant–microbe interactions, for subsequent high-resolution EM studies.

We have demonstrated that lab-based XRM provides a unique means of 3D imaging larger and more complex plant samples than is typically possible for light and electron microscopy, providing researchers with an opportunity to visualize important features in high resolution over a wide range of magnifications within the context of larger tissues, organs, or whole plants. The use of XRM to augment volume electron microscopy by providing a high-resolution 3D roadmap makes multiscale correlative imaging a powerful option for visualization of sub-cellular and cellular details, again co-registered across scales and within a larger sample context. We hope that describing numerous examples of robust, high-resolution 3D imaging strategies from a wide range of plant samples will encourage more researchers to explore lab-based XRM as a powerful and essential tool for understanding cell, tissue, and organismal plant biology, as well as assist technical staff in imaging facilities who may not have experience in XRM imaging of plant samples.

The following materials are available in the online version of this article.

Supplemental Figure S1 . Sample preparation workflow for XRM.

Supplemental Figure S2 . Tissue level segmentation of Setaria viridis spikelets and bristles.

Supplemental Figure S3 . Inflorescence development.

Supplemental Figure S4 . Multiscale imaging.

Supplemental Figure S5 . Identification and 3D volume measurement of individual cells from XRM scan of maize root tip.

Supplemental Figure S6 . Measurement data from 3D segmentation of maize root tip XRM scan.

Supplemental Figure S7 . Multiscale in situ imaging of host–microbe interaction.

Supplemental Figure S8 . Seed morphology.

Supplemental Table S1. Preparation details for all samples illustrated in figures.

Supplemental Table S2 . X-ray microscope scan parameters for images shown in figures.