Quantitative characterization of bamboo cortex structure based on X-ray microtomography

Bamboo is a natural fiber composite with layered structure. Millions of years of evolution have endowed bamboo with the most effective structure in nature. The ingenious microstructure provides bamboo with excellent mechanical properties. Bamboo culm is composed of the cortex, a middle layer, and a pith ring. The cortex refers to the area starting from the periphery of the culm wall to the vascular bundles. The present study obtained the two-dimensional (2D) microstructure of bamboo cortex cells by optical microscopy and characterized the three-dimensional (3D) structure through high-resolution X-ray microtomography (µCT). The 2D anatomical parameters of cortex cells were measured to verify the reliability of cortical cell classification in 3D models. Based on the analysis, the bamboo cortex cells were classified into four layers: epidermis layer, hypodermis layer, transitional layer, and parenchyma layer. The total volume of the reconstructed area of interest in μCT was 7.85 × 106 µm3, in which the total pore volume of bamboo cortex was 2.84 × 106 µm3, the average pore volume of bamboo cortex was about 1.54 × 103 µm3. Thus the porosity was 36.1%, and the relative density was 0.639. Studies on bamboo anatomical structure, especially three-dimensional digital characterization, will enrich the bamboo microstructure database. Besides, the three-dimensional structure of the bamboo cortex revealed in this study can provide a reference for optimizing composite material hierarchy and biomimetic design.


Introduction
Bamboo is a natural, functionally graded composite material widely distributed in the tropical, subtropical, and temperate regions (Nogata et al. 1995;Amada et al. 1996). Bamboo culms and woven mats have been used for millennia in traditional construction (Jiang 2007;Lucas 2013). The excellent flexibility 1 3 Vol:. (1234567890) (Fang et al. 2018) and damage tolerance (Low et al. 2006) properties of bamboo are closely related to its microstructural features. Millions of years of evolutionary optimization have endowed bamboo its ingenious structure, with excellent mechanical properties and the ability to adapt to environmental challenges. The bamboo culm wall is solid part of bamboo culm which consists of three layers, called the inner, middle, and outer layers. The outer layer called the cortex presents a watertight seal to prevent moisture loss from the living culm (Liese et al. 2015). Its middle layer, which is the main component, is composed of optimized unidirectional fiber-reinforced composites with cellular parenchyma as the matrix (Habibi et al. 2015). The fibers are the major components that provide the strength and toughness of bamboo (Lo et al. 2004;Zou et al. 2009). Finally, the inner part of the culm wall called pith ring is composed of thickwalled stone cells. Pith cavity surrounded by stone cells is critical for bamboo to overcome bending force during its fast growth (Guo et al. 2019).
Researchers have expounded on the structure of bamboo culm over the years. These studies mainly focused on the fibers, parenchyma cells and vascular bundle (Chen et al. 2020a, b;Song et al. 2017;Chen et al. 2019;Chen et al. 2020a, b;Dixon et al. 2014;Lian et al. 2020;Palombini et al. 2016;Palombini et al. 2020a, b;Wu et al. 2021). Besides, a few studies analyzed the cortex of bamboo. Liese et al. reported that the bamboo cortex contains epidermis and hypodermis, and the primary function of bamboo cortex is to block water and protect the tissues (Liese et al. 1998). Meanwhile, Wang et al. found that the contact angle of the untreated bamboo surface was 114°, indicating a hydrophobic surface (Wang et al. 2020). Cui noted that the silica in the bamboo cortex forms a perfect bonding interface to cellulose fibers. Moreover, the presence of silica significantly increased the critical stress of random cracks in the bamboo cortex (Cui et al. 2021). Bamboo in its natural habitat acts as a cantilever beam with a fixed support in the earth and subjected to its own weight and wind load. Therefore, it has a naturally optimized structure to resist bending moments. The strengths are highest along the outside and lowest in the inside surfaces. (Tan et al. 2011). The features of the cortex also assist in bamboo identification (Ghosh et al. 1960).
Studies on the structure of bamboo provide details on the geometric and material parameters necessary to understand its mechanical and biological functions better. However, studies on the bamboo cortex have been mainly on two-dimensional characterization due to the lack of three-dimensional analysis of structural features. X-ray microtomography (µCT), one of the most powerful techniques used for the three-dimensional analysis of material microstructure due to high resolution and non-invasive features (Landis et al. 2010), has been used in wood and bamboo structure. Dixon et al. (2018) used µCT to collect bamboo parenchyma tissue images and 3D printing technology to study its mechanical properties. Palombini (2020) obtained the shape and parameters of bamboo internode parenchyma cells (Fig. 1J) and vascular bundles in the bamboo node (Figs. 1D and 1G) (Palombini et al. 2016). Wu et al. (2021) reconstructed the bamboo vascular bundles (Fig. 1K) and parenchyma cells in the bamboo internode and calculated the area and diameter of the parenchyma cells. In addition, some scholars used X-ray computed tomography (CT) to measure the density distribution of Moso bamboo (Huang et al. 2015). These researches proved that X-ray microtomography is an effective method to obtain the three-dimensional anatomical structure of bamboo. However, no studies have used this technique for the three-dimensional analysis of bamboo cortex.
The present study analyzed the structure of bamboo cortex cells by optical sectioning and highresolution CT. The bamboo cortex cell parameters, including a three-dimensional model and the structural features, were noninvasively obtained by X-ray micro-computed tomography (µCT). The study's findings will improve our understanding of the anatomical structure of bamboo cortex cells, and explain the relationship between the anatomical structure and the mechanical and growth of bamboo.

Sample preparation
Four-year-old bamboo plants (Phyllostachys heterocycla) were obtained from Taiping, Anhui Province, China. In order to avoid the variability of bamboo characteristics, the tenth bamboo tube was taken as the experimental material from bottom to top (Liu. 2017) ( Fig. 2A and B). The sample used to obtain the three-dimensional microstructure of bamboo cortex was first air-dried at a temperature of 23 ℃ and relative humidity of 46% and then processed into small pieces of about 2 mm × 2 mm × 2 mm (L × R × T; Fig. 2D). These samples used to obtain bamboo cortex slices were softened in hot water (60 °C) for 5 days (Fig. 2F). Figures 2H and I were the radial section and cross-section slices with a thickness of 10-20 µm obtained by flat push slicer (Leica SM2000R). The slices were observed, and images were captured under Light microscope (LIOO S600T). The length (L) and width (W) of the cell lumen and the thickness of the wall were measured using the cross-section and radial section slices, respectively. Measurement of each parameter was repeated 50 times, and the values were represented as an average. Anatomical features of the cortex cells were measured in ImageJ software (http:// rsb. info. nih. gov/ ij).

X-ray microtomography (μCT)
The samples were individually scanned via highresolution microtomography (Bruker SKYSCAN 2214, Bruker, Germany). The sample is fixed on the cylindrical sample holder provided with the instrument (As is shown in Fig. 2D the red wireframe). Due to the small size, the bamboo cortex sample was entirely within the field of view. The sample was then scanned with an X-ray source at a voltage of 55 kV and a current of 160 µA, resulting in a voxel size of 0.8 µm after binning. A total of 1217 projections were obtained from the sample by the CCD camera;  (Palombini et al. 2020a, b), C Nodal region of the apical stem (Palombini et al. 2020a, b), D Three-dimensional CT reconstruction of vascular bundles of the nodal bamboo region with secondary branching (Palombini et al. 2020a, b), E Bamboo node, F Radial section of the bamboo node; G Three-dimen-sional μCT reconstruction of node (Palombini et al. 2016), H Bamboo internode, I Sample used to obtain the three-dimensional structure of parenchyma and vascular bundle in bamboo internode, J Three-dimensional μCT reconstruction of bamboo parenchyma (Palombini et al. 2020a, b), K Three-dimensional μCT reconstruction of the vascular bundle (Wu et al. 2021) The images were combined in the software (DATA-VIEWER) provided by the manufacturer to obtain the gray image in BMP format.

Reconstruction and data analysis
The open-source FIJI/ImageJ software was used to select the bamboo cortex as the region of interest (Fig. 3A). Then, the selected region was imported into Avizo software (FEI, Hillsboro, Oregon, USA), and the images were processed. Figure 3 shows the processing of the image stack in Avizo software. The gray-scale slice image of the region of interest was first imported (Fig. 3B) into the software, and the brightness and contrast were adjusted (Fig. 3C). Then, image denoising was done using the non-local means filter (Fig. 3D), and the interactive threshold segmentation (Fig. 3E) and hand segmentation ( Fig. 3F) were carried out. Figure 2E was the 3D reconstruction model of bamboo cortex.
Cellulose and other carbon-based compounds readily absorb X-rays, and the difference in X-ray attenuation between plant tissue and the surrounding air provides excellent contrast (Brodersen 2013). Therefore, the absorption of X-ray by bamboo cell wall and cell cavity is obviously different, the gray values presented by the cell wall and the cell cavity were significantly different. Interactive threshold segmentation was used to separate the pores from the cell wall. A few cells that failed to achieve ideal segmentation were separated by hand segmentation. Using the Brush tool in Avizo software to manually segment these cell cavities (As shown in red wireframe areas in Figs. 3D, E and F). The segmented image was used to obtain the three-dimensional imaging and related data collection in Avizo. The cortex cell volume, surface area, slenderness ratio, and sphericity were calculated using the Avizo software. Here, length3d (L) represents the maximum Feret diameters of the cells, and breadth3d (B) represents the largest distance between two parallel lines touching the cell without intersecting it and lying in a plane orthogonal to the maximum three-dimensional Feret diameter. The volume of the selected region (V Ri ) was determined using Eq. (1) (Palombini et al. 2016;Zellnig et al. 2002): where L x is the voxel size, i indicates each section, n is the total number of sections, and a R i is the area of a specific region.
(2) (Palombini et al. 2016; Gibson et al. 2003): where V V is the volume of pores in the selected region is the relative density of the selected region, * is the density of the cellular material, and s is the density of the solid from which the cell walls were prepared. The sphericity of the cells was determined using Eq. (3) (Singh et al. 2016;Jiang et al. 2011) where V P c and A P c indicate the volume and the surface area of the cortex cells, respectively. As ψ approaches 1, the cell becomes spherical. (1)

Results and discussion
Characteristics of cortex microstructure The microstructure of bamboo cortex was characterized by Avizo, and the overall porosity of the cortex was obtained. Through optical microscope, cells of various shapes were found in the cross-sectional and radial sections of the cortex. Figures 2H and I show the cell morphology of the bamboo cortex in the sections under an optical microscope. Compared to the two-dimensional information of the slice (Liese et al. 2015), the three-dimensional images reconstructed by Avizo software intuitively showed the morphological differences of cortical cells (Fig. 4A), indicating that μCT was an effective means to study cortex cells. The total volume V R of the reconstructed area of interest in μCT was 7.85 × 10 -3 mm 3 , in which the volume V V of the pores was 2.84 × 10 -3 mm 3 . Therefore, the porosity φ of bamboo cortex was about 36.1% (Table 1), while according to Palombini et al., the porosity of bamboo parenchyma cells in middle part of bamboo culm was 72.6% (Palombini et al. 2016). The porosity of the bamboo cortex is much smaller than that of the parenchyma cells in middle part of bamboo culm. The relative density of bamboo cortex cells calculated using Eq. (2) was 0.639, which is between those of the parenchyma cells (0.274) and fiber (0.907) (Palombini et al. 2016). These findings indicate that compared with the parenchymal cells in the ground tissue, the cortex cells have lower porosity and a higher cell wall material content and subsequently a relatively high density. Meanwhile, pores, screened by pore counting plug-in, had an average volume of about 1.54 × 10 -6 mm 3 in the cortex. The strength is the highest along the outside surface when bamboo is bent under stress (Tan et al. 2011). The characteristics of small pores and high density of bamboo cortex may give it unique mechanical properties, which will not be easily destroyed when subjected to external forces.

Cell type of bamboo cortex
Previous studies divided the cortex cells into the epidermis layer, hypodermis layer, and parenchyma cells (Liese et al. 1998). However, in the present study according to the bamboo cortex slices and 3D reconstruction model characteristics, bamboo cortex cells were classified into four types, including the epidermis layer (Fig. 4B), hypodermis layer (Fig. 4C), transitional layer (Fig. 4D), and parenchyma layer (Fig. 4E) cells. The outermost layer was the epidermis layer consisting of one layer cells, followed by the hypodermis layer consisting of one or two layers of cells, and then the transitional layer about 2-4 layers and finally the parenchyma layer which has a variable number of layers. Only slight differences were detected in the morphology between the transitional layer cells and parenchyma cells, which may explain why previous studies divided cortex cells only into three categories (Liese et al. 2015). Transitional layer cells and parenchyma cells are easily distinguished by the three-dimensional topography ( Fig. 4D and E), with cells in the transitional layer being flat while the parenchyma cells in the cortex are columnar. Figure 4 shows the 3d reconstruction model features of bamboo cortex cell cavities from the outer side to the inner side in Avizo, revealing the epidermis cells as the first layer (Fig. 4B), hypodermis cells as the second layer (Fig. 4C), transitional cells as the third layer (Fig. 4D), and the parenchyma cells as the  fourth layer (Fig. 4E). The cells of the epidermis are similar to peanuts and are short columnar, the cells of the hypodermis layer are elongated columns, the cells of the transitional layer are flat, and the parenchyma cells of the cortex are also columnar but have the largest volume. Typical cell models for each layer are at the right of Fig. 4. Table 2 shows the volume ( V P c ), area ( A P c ), sphericity (ψ), length3d (L), and breadth3d (B) of cells in each layer of the cortex. In order to verify the reliability of the classification of cortex cells in the 3D model, 2D anatomical parameters of cortex cells were measured quantitatively, such as length, width and double cell wall thickness. Figure 5 shows the 2D anatomical parameters measurement results of cell parameters in radial section slices (Fig. 5B) and cross section slices (Fig. 5C). The 2D anatomical parameters of cortex cells showed that it was reasonable to divide the cortex into four types according to 3D model. In the radial section, the transitional layer cells are longer than the parenchyma cell layer, and the width is slightly smaller than that of the parenchyma cells (Fig. 5A); In the cross section, the transition layer cells are approximately elliptical, while the parenchyma cells are nearly circular (Fig. 5A).

Epidermis layer
The epidermis layer had one layer of cells arranged in parallel along the axial direction at the periphery. The long axis of the epidermis layer cells was perpendicular to the axis of the bamboo culm. Most of the cells were short columnar, while a few were irregular (Fig. 4B). Liese's anatomical analysis had indicated that these irregularly shaped cells could be siliceous cells and embolus cells (Liese et al. 1998). Siliceous cells contain silica, which makes the epidermis layer denser than the other cortex cells, the presence of silica significantly increased the critical stress of random cracks in the bamboo cortex (Cui et al. 2021).
The average volume of the epidermis layer cells was 917.81 µm 3 , the average sphericity was 0.85, and the average length3d-breadth3d ratio was 1.37. In the cross-section, the average length of epidermis layer cells was 11.15 µm, the average width was 5.03 µm, and the mean double wall thickness was 3.78 µm.
In the radial section, the average length of epidermis layer cells was 10.65 µm, the average width was 5.6 µm, and the mean double wall thickness of cells was 3.61 µm.

Hypodermis layer
The hypodermis layer had 1-2 layers of long columnar cells with a small lumen and an obvious cell wall thickening (Fig. 4C). The long axis of the cell was parallel to the axis of the bamboo culm. The cells were nearly round in the cross-section and closely arranged. The average volume of the hypodermis layer cells was 714.22 µm 3 , the average sphericity was 0.76, and the average length3d-breadth3d ratio was 2.74. In the cross-sectional slices, the average length of hypodermis layer cells was 5.27 µm, the average width was 5.13 µm, and the mean double wall thickness of cells was 4.29 µm. Meanwhile, in the radial section, the average length of the epidermis layer cells was 18.1 µm, the average width was 4.49 µm, and the mean double wall thickness of cells was 4.44 µm.

Transitional layer
The transitional layer had 2-4 layers of flat cells (Fig. 4D), whose cross-section appeared almost elliptic, with increasing cell volume from outside to inside. The long axis of the transitional layer cells was parallel to the axis of the bamboo culm. The average volume of the Transitional layer cells was 1258.19 µm 3 , the average sphericity was 0.75, and the average length3d-breadth3d ratio was 1.74. In the cross-sectional slices, the average length of the transitional layer cells was 12.74 µm, the average width was 9.17 µm, and the mean double wall thickness of cells was 4.6 µm. In the radial section, the average length of epidermis layer cells was 32.15 µm, the average width was 9.34 µm, and the mean double wall thickness of cells was 5.54 µm.

Parenchyma layer
The transitional layer was followed by the parenchyma cells (Fig. 4E), in contact with vascular bundles. However, due to the uneven distribution and arrangement of vascular bundles, the number of layers of cortex parenchyma cells varied. The long axis of the parenchyma cells was parallel to the axis of the bamboo culm. Cells in the parenchyma layer were columnar and larger than those in the hypodermis layer, with an average volume of 3.12 × 10 -6 mm 3 , sphericity of 0.78, an average diameter of 17.32 µm, and an average length3d-breadth3d ratio of 2.1. Meanwhile, Palombini et al. (2016) reported an average volume of 6.65 × 10 −6 mm 3 , sphericity of 0.75, and an average diameter of 25.83 µm for the parenchyma cells. These results indicate that the cortex parenchyma cells were similar to the ground tissue of bamboo but much smaller in size. In the cross-sectional slices, the average length of parenchyma cells was 14.06 µm, the average width was 10.99 µm, and the mean double wall thickness was 5.94 µm. In the radial section, the average length of the parenchyma cells was 23.8 µm, the average width was 11.15 µm, and the mean double wall thickness of cells was 7.45 µm.
Volume and sphericity Figures 6A, B, C, and D show the relationship between the volume of the four layers of cells and their sphericity, with each dot representing an analyzed cortex cell generated by the particle analyzer plug-in. The point cloud from Origin 9.0 (Origin Lab Corporation, UK) showed the overall relationship between cell volume and sphericity. The cell volumes of epidermis and hypodermis layer were mostly below 2000 µm 3 , and that of the transitional layer mostly below 4000 µm 3 . Meanwhile, the volume range of cortex parenchyma cells varied from 0 to 10,000 µm 3 (Fig. 6D). The point cloud distribution indicated the highest sphericity for the epidermal cells ( Fig. 6A-a), indicating a nearly spherical cell cavity. The sphericity of the hypodermis layer cells, transitional layer cells and parenchyma cells was slightly lower, consistent with the cell cavity morphology generated via CT reconstruction ( Fig. 6B-b,  Fig. 6C-c and Fig. 6D-d). From Fig. 6, the relationship between sphericity and cell volume is not obvious. In Table 2, the volume of the four-layer cells showed an increasing-decreasing-increasing-increasing trend from outside to inside, while the sphericity showed a decreasing-decreasing-decreasing-increasing trend from outside to inside.

Porosity
The three-dimensional model was reconstructed by stacking slices and the thickness of each slice was 0.8 µm. The porosity of each slice is different. The overall porosity of bamboo cortex (ψ = 36.1%) does not change due to the change in evaluation direction. However, the evaluation of bamboo cortex porosity in radial and axial-section can reveal the changes of cortex porosity in direction. The longitudinal direction (L) of the 3D model consists of 400 slices with an average porosity of 36.1% and a standard deviation of 1.36; the tangential direction consists of 188 slices with an average porosity of 36.1% and a standard deviation of 7.27. The porosity of the bamboo cortex did not change significantly along the length in longitudinal direction (L) (Fig. 7A). While, the porosity of bamboo cortex samples fluctuated greatly along the tangential direction (T) (Fig. 7B). The porosity of the slices fluctuated obviously in the parenchyma cell layer hypodermis cell and epidermis layer, while it fluctuated little in the transitional layer. When the proportion of cell cavities in the slice is large, the porosity of the slice will be large ( Fig. 7B-a and c). When the proportion of cell cavities in the slice is small and the proportion of cell walls is large, the porosity of the slice will be small ( Fig. 7B-d). The porosity difference of each section of the transitional layer is small, which may be due to the irregular arrangement of cells and small cell lumen ( Fig. 7B-b).

Conclusion
Three-dimensional characterization of bamboo cortex cell structure via high-resolution computed tomography was feasible in the present study. Furthermore, the volume ( V P c ), area ( A P c ), sphericity (ψ), length3d (L), breadth3d (W) and other parameters of bamboo cortex cells were obtained, and bamboo cortex cells were classified into four types based on morphological features. The cortex cells included the epidermis cells, hypodermis cells, transitional layer cells, and parenchyma cells from the outside to the inside. The analysis revealed a closedcell foam material for bamboo cortex cells similar to parenchyma cells in the middle part of bamboo culm. These findings increased our understanding of the cell structure of bamboo cortex and help understand the corresponding mechanical and biological characteristics of bamboo at different scales. The ingenious four-layer structure of bamboo cortex may be the key to its ability to absorb and transfer stress. Meanwhile, the study provided theoretical and data basis for future improvement of bamboo production and utilization. There are some defects in the segmentation of bamboo cortex cell wall and cell lumen by threshold value in this study, which may cause some deviation of data. Therefore, developing a suitable method for bamboo cell segmentation is a potential research direction. Besides, the mechanism of bamboo cortex absorption and transfer stress remains to be further studied, for instance, nanoindentation technology could elucidate the elastic/plastic properties of bamboo cortex cell wall. The model of bamboo cortex obtained by µCT can be used for finite element analysis to simulate the flexural behavior of the culm and obtain the role of the cortex in this process.
Funding The authors gratefully acknowledge financial support from the National Natural Science Foundation of China (32071856, 32101608).

Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References
Amada S, Munekata T, Nagase Y et al (1996) The mechanical structures of bamboos in viewpoint of functionally