SEARCH

SEARCH BY CITATION

Keywords:

  • basal root;
  • Cicer arietinum;
  • image analysis;
  • lateral root;
  • Phaseolus vulgaris;
  • root growth;
  • root kinematics

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Quantification of overall growth and local growth zones in root system development is key to understanding the biology of plant growth, and thus to exploring the effects of environmental, genotypic and mutational variations on plant development and productivity.
  • We introduce a methodology for analyzing growth patterns of plant roots from two-dimensional time series images, treating them as a spatio-temporal three-dimensional (3D) image volume. The roots are segmented from the images and then two types of analysis are performed: 3D spatio-temporal reconstruction analysis for simultaneous assessment of initiation and growth of multiple roots; and spatio-temporal pixel intensity analysis along root midlines for quantification of the growth zones.
  • The test measurements show simultaneous emergence of basal roots but sequential emergence of lateral roots in Phaseolus vulgaris, while lateral roots of Cicer arietinum emerge in a rhythmic pattern. Local growth analysis reveals multimodal transient growth zone in basal roots. At the initial stages after emergence, the roots oscillate rapidly, which slows down with time.
  • The methodology presented here allows detailed characterization of the phenomenology of roots, providing valuable information of spatio-temporal development, with applications in a wide range of growing plant organs.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

The root system architecture is central to a plant’s ability to survive, grow, and produce yield as it impacts key physiological processes such as nutrient and water uptake, anchorage, and inter- and intra-plant competition (Osmont et al., 2007). As a seedling grows to become a mature plant, root architecture develops continuously to form the final root system. Although the spatial distribution of the roots of a mature plant may seem to be most important for a plant’s productivity, the developmental pattern of the roots at each stage is also equally important as it controls the development of the whole plant in the subsequent stages. Therefore, a spatio-temporal description of root development is critical for understanding plant growth. As the overall growth and development of roots are a result of cumulative effects of local growth, it is also important to study the growth patterns of the roots at multiple scales in space and time to explore both overall development and local growth zones. Such multiscale spatio-temporal descriptions of root development provide deeper insights into the biology of plant roots and their interactions with the environment (Walter et al., 2009).

In recent years, time-lapse imaging coupled with semi-automated analysis tools have been developed to assess various kinematic parameters of root growth. Among these, thresholding (Miller et al., 2007; Yazdanbakhsh & Fisahn, 2010), skeletonization (Armengaud et al., 2009) and computer tracking (French et al., 2009) based image analysis methodologies provide assessment of overall root growth. However, such measurements are not suited for analysis of local growth zones of the growing roots. Optical-flow based methods (van der Weele et al., 2003; Basu et al., 2007; Chavarria-Krauser et al., 2008), in contrast, allow assessment of local growth zones of the roots. But the limitation of both approaches is that they remain mutually exclusive as the former requires uniformly colored roots while the latter requires inherent or externally added patterns on the roots. In addition, these approaches are also unable to analyze root initiation and branching, leaving the temporal description of root system development incomplete.

We therefore designed a new methodology to explore the development of roots from emergence until the later stages, with or without patterns on the root images. Thus, the new methodology represents a unified approach with additional features for exploring the finer details of root development and growth, which are unavailable in any previous method (van der Weele et al., 2003; Basu et al., 2007; Miller et al., 2007; Chavarria-Krauser et al., 2008; Armengaud et al., 2009; French et al., 2009; Yazdanbakhsh & Fisahn, 2010). The new methodology treats a stack of two-dimensional (2D) time-lapse images as a three-dimensional (3D) volume of image data, allowing 3D image analysis and reconstruction tools to be used for exploration of growth patterns of the roots in space and time, albeit with specific customizations and newer interpretations. The proof-of-principle of the technique was tested by analyzing initiation and growth of basal roots in rajmash bean (an Indian cultivar of common bean, Phaseolus vulgaris), and lateral roots in both chickpea (Cicer arietinum) and rajmash bean. The comparison of growth patterns of basal and lateral roots, and lateral roots of two different plants provides intriguing insights into the development of the root systems of these two legumes.

Materials and Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Plant growth

The new methodology uses 2D images of growing roots obtained from plants grown in germination paper sandwiched between a glass sheet at the front and a plastic sheet at the back (Fig. 1a). We used seeds of chickpea (Cicer arietinum L.) cultivated variety DCP 92-3 developed by the Indian Institute of Pulses Research (IIPR), India and rajmash bean (Phaseolus vulgaris L.) germplasm EC541702 collected from IIPR. Seeds were surface-sterilized with 6% sodium hypochlorite solution for 5 min and rinsed thoroughly with distilled water. Seeds were germinated at 25°C in darkness inside the growth chamber (Acm-78094 S; ACMAS Technology, New Delhi, India) for 48 h in brown germination paper (Anchor Paper Co., St. Paul, MN, USA) moistened with nutrient solution composed of (in μM) 3000 KNO3, 2000 Ca(NO3)2, 1000 NH4H2PO4, 250 MgSO4, 25 KCl, 12.5 H3BO3, 1 MnSO4, 1 ZnSO4, 0.25 CuSO4, 0.25 (NH4)6Mo7O24 and 25 Fe-NA-EDTA. Germinated seeds of C. arietinum with 5–6-cm-long primary roots and those of P. vulgaris with 2–3-cm-long primary roots were transferred to a sheet of blue germination paper (Anchor Paper Co.) stiffened by attaching a glass sheet to stabilize the root system (Fig. 1a). The bottom 2–3 cm of the blue germination paper containing the seedling was immersed in nutrient solution and the entire set-up was placed inside the growth chamber at a temperature of 22 ± 1°C for C. arietinum and 25 ± 1°C for P. vulgaris seedlings. To ensure that the roots grow in the dark, the set-up is placed in a growth chamber, divided into two compartments, with the shoot growing in the upper, bright (12 h : 12 h, day : night cycle) and the roots in lower, dark environment (Fig. 1b).

image

Figure 1. Experimental set-up for the in vivo root imaging and analysis methodology. (a) Schematic of the transparent growth system where the seedling roots grow on a germination paper placed between a clear glass sheet at the front and a plastic sheet at the back, enforcing nearly planar root architecture. (b) The transparent growth system is placed in the plant growth chamber divided into two compartments. The lower compartment contains a camera on a tripod which is tethered to a computer for time-lapse imaging, and flashes. It is kept dark by covering with a thick black cloth. The shoot appears in the upper chamber through the cloth and is exposed to light (12 h : 12 h, day : night cycle). (c) A sample image of a growing bean (Phaseolus vulgaris) seedling with basal roots B1–B6 and lateral roots L1 and L2 labeled (root labels B1–B6 and L1–L2 have no physiological meaning.). The black spots on the root were added by sprinkling graphite particles onto the root to provide patterns for local growth estimation. (d) A stack of nine time series images is shown from a sequence of 81 images. Slicing planes P1–P5 are passed through the stack to obtain spatio-temporal image sections on which the roots are segmented (cyan contours). Red dots show the intersection of the contours with the images and are used as seed points. (e) An example sliced image from plane P3. (f) Using the seed points, root images are segmented (green contours). (g) From the edge contours of the root images, a spatio-temporal 3D structure is constructed. The thin green lines enveloping the spatio-temporal 3D structure show edge contours of all 81 images. Labels show the structures corresponding to roots in (c). (h) Spatio-temporal patterns of graphite particles along the root midlines.

Download figure to PowerPoint

Image acquisition

Roots were imaged with a Nikon D200 digital camera. The camera was connected to a computer through which time-lapse imaging was performed. We used two external flashes (Nikon SB-800 and Nikon SB-600) to capture the images to minimize exposure of the roots to light (Fig. 1b). The external flashes were triggered wirelessly by the built-in flash of the camera. Light from the flashes was directed away from the roots to avoid specular reflection on the roots, which can cause unwanted patterns.

One day after transferring the seedlings to the growth pouch, the roots were photographed for up to 5 d at 30-min intervals to assess total growth of the basal and lateral roots. For uniformly colored roots of legumes, graphite particles were sprinkled on the roots to add patterns for analysis of local growth zones (Fig. 1c). Roots were photographed at 5–10-min intervals for up to 8 h (Fig. 1c). While the set-up allows continuous imaging of the roots for relatively long periods of time, up to 1 wk or more, the assessment of local growth zones using added patterns of graphite particles is limited by time, because growth of the roots makes the graphite particles too sparse to analyze, limiting the study to a maximum of 8 h.

Image analysis

The steps for the image analysis are described in the following section, with additional mathematical details presented in Supporting Information Notes S1 together with the flow chart (Fig. S1).

Image preprocessing and stabilization

Images captured with a Nikon D200 digital camera have a high resolution (10 Megapixels), making it difficult to analyze a number of these images simultaneously. Therefore, before proceeding with the analysis, the program allows cropping of the images to isolate a specific group of roots for analysis. In addition, to reduce the resolution of the images so that a large number of images can be analyzed together, the program also allows down-sampling of the images.

Images captured in the growth chamber are susceptible to a small amount of vibrational motion as a consequence of the ventilating fans in the growth chamber, which is amplified in close-up images (Supporting Information Video S1). If uncorrected, these vibrations lead to erroneous results for root trajectory. Therefore, before beginning processing of the images for extraction of dynamical data of root growth, identifiable marker points on the background or on the nongrowing parts of the roots, for example, ink marks on the germination paper, are tracked among all the images using the block matching technique (Scarano, 2002; Basu et al., 2007). The points for tracking are chosen by the user on one representative reference image, A. From n such points, displacements inline image are calculated where inline imageand inline image are coordinates of the ith point in the reference image A and current image B. The median value of displacements inline image is the shift of the image B as a result of vibrations, which is subtracted from the pixel coordinates to eliminate vibration and ‘stabilize’ the images (Fig. S2, Video S2). Further details of the vibration stabilization procedure are provided in Notes S1.

Spatio-temporal image slicing

After stabilization of the images, the first step in the analysis is segmentation of the roots. Uniformly colored roots with good contrast relative to the background can be segmented by choosing a threshold value of image intensity (Miller et al., 2007). However, in roots with marker points, thresholding cannot segment them as the patterns also become separated. Although many semi-automated image segmentation methodologies exist (Canny, 1986; Kass et al., 1987; Barrett & Mortensen, 1997), applying these to a large set of images individually is very time-consuming. Therefore, we developed a methodology where the entire stack of images is first sliced with a number of user-specified spatio-temporal planes (P1–P5 in Fig. 1d) such that these planes intersect the roots in almost all images. Image data are interpolated by trilinear interpolation to generate spatio-temporal images of the roots on these slicing planes. Fig. 1(e) shows the sliced and segmented image of plane P3.

Segmentation by livewire

From the interpolated spatio-temporal images on the slicing planes, the roots are segmented semi-automatically using the ‘livewire’ algorithm (Barrett & Mortensen, 1997; Hamarneh et al., 2005). The points of intersection of the segmented contours on the sliced planes and the original image planes are treated as the seed points. The livewire algorithm finds the minimum cost path between two seed points (Fig. S3). The cost of the path is determined from image features such as intensity gradient magnitude, intensity gradient direction, laplacian zero crossing, and Canny edge. After the seed points are established, the minimum cumulative cost between two seed points is calculated using dynamic programming following the optimal graph search method (Dijkstra, 1959). The segmentation of roots by the livewire algorithm is completely automated and does not require any user intervention after the seed points have been established. Edges obtained are smoothed by applying a Gaussian filter. The user can choose the amount of smoothing by choosing the size of the filter kernel. Following segmentation of the roots in the interpolated images, edge contours are generated (cyan lines in Fig. 1d) which intersect the original images. Treating these points of intersection as ‘seeds’ (red dots in Fig. 1d,f), we segment the roots automatically using livewire, resulting in edge contours of the roots on all original 2D images (green contours in Fig. 1f).

Identification of tip and midlines of the roots

From the segmented and smoothed edge contour of a root, the user identifies a point near the root tip on one of the time series images as an initial guess for the root tip. The location of the highest curvature point on the edge contour in the neighborhood of the selected point is finalized as the root tip (blue dots in Fig. S4). The neighborhood used in the method is ± 10% of the total length of the root edge. Then the point on the edge contour of the subsequent image which is nearest to the current root tip is identified as the initial guess for the root tip in that image. This step is followed by the same algorithm to relocate the tip to the highest curvature position. In this way the root tip is automatically identified in all the images. In case the algorithm fails to identify the root tip accurately, the user can manually reposition it. After the root tip is identified, the root base is identified at the point of emergence (yellow dots in Fig. S4). As typically the base point is static, it is copied in all the images. In case the base point of the root moves, the user has to manually reposition it.

For every pixel within the edge contour of the root, the Euclidean distance transform (EDT) is calculated (Fig. S4a). EDT is the distance of a pixel from the nearest edge point. The trajectory of the maximum values of EDT is the root midline (Fig. S4b). This line is obtained by initiating a search at the root tip with a direction vector pointing to the root base and using the procedure described by Miller et al. (2007). Because the EDT is calculated at pixel resolution, the detected midline is also at pixel resolution.

Construction of spatio-temporal 3D structures

Following segmentation, two types of analysis are performed to study overall root growth (Fig. 1g) and local growth zones (Fig. 1h). First, from the segmented edge contours an optimal field function φ(x, y, t) is calculated such that on the surface of the spatio-temporal 3D structure φ(x, y, t) = 0 (Cong & Parvin, 1999). Here (x, y) are image plane coordinates and t is time. The isosurface of this function at φ(x, y, t) = 0 is the spatio-temporal 3D structure (Fig. 1g).

The second type of analysis uses spatio-temporal variations in pixel intensities along the root midline (Fig. 1h). The pixel intensities along the root midline are interpolated from the images using bi-linear interpolation and expressed as J (s, t), where s(x, y) is the distance of point (x, y) from the root base along the midline. As small changes in the root midline can cause substantial changes in the interpolated intensity function J (s, t), for every spatio-temporal point, pixel intensities are averaged within a neighborhood of three pixels to obtain an average value of J (s, t). When J (s, t) is displayed as J (x, y, t) it provides the spatio-temporal view of the changes in the pixel intensities along the midline of the roots and illustrates local growth zones (Fig. 1h) (Erickson & Sax, 1956).

Analysis of root growth

The length of the root midline is the root length. The striped patterns on the spatio-temporal map of the pixel intensities along the root midline show the local growth zones (Fig. 1h). Treating these patterns similar to fluid flow, the spatio-temporal displacement vectors u = (us, ut) are obtained using the block matching technique (Scarano, 2002; Basu et al., 2007) (Fig. S5a–c). Here us and ut are the components of the displacement vector along space and time, respectively. The temporal component of the displacement vector ut is the time gap between two consecutive images so that the spatial component us is the displacement of a marker position between those two images. Therefore, us/ut quantifies the local tissue velocity, that is, the local growth velocity with respect to the root base. From the displacement vectors we calculate the streak lines. For any point x0 = (s0, t0) on the spatio-temporal midline intensity map, the next location can be calculated as x1 = x0 + u (i.e. s1 = s0 + us and t1 = t0 + ut). Thus, by joining n such points, x0, x1, ..., xn, the trajectory of a marker point is obtained which is the streak line (Fig. S5d–f). Root growth velocities are interpolated along the streak lines and smoothed using overlapping polynomials before calculation of the relative elemental growth rate, inline image

Validation

To assess the accuracy of the image analysis methodology we compared results obtained from artificially generated sequences of 41 root images (resolution 300 × 600 pixels). The artificial image sequences were generated from pre-designed theoretical growth zones with specific REGR distributions – unimodal and bifurcating. In both cases, the peak REGR values were 0.05 per time-point. Examples of these artificial image sequences are presented in Videos S3 and S4. In the first example, the artificial root has a unimodal growth zone defined by a Gaussian REGR which results in uniform elongation of the root. In the second example, the unimodal elongation zone bifurcates into double-peaked elongation zones defined by Gaussian distributions of REGR. The artificial root images were marked with 35 black dots so that the local growth zones can be estimated. Root- mean-square-deviations (RMSDs) of the measured growth parameters from the theoretical values were calculated to quantify the accuracy of the image analysis system.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Analysis of spatio-temporal growth of the root system

A spatio-temporal 3D structure was constructed from the edge contours as shown in Fig. 1(g). As the roots grow, the spatio-temporal 3D structure widens along the time axis. Therefore, its slope along the tip (i.e. the rate of widening with time) of each root shows variations in growth rate with time. Furthermore, the geometry of the spatio-temporal 3D structure illustrates the overall growth patterns; for example, a wavy structure (basal root B2 in Fig. 1g) indicates oscillatory root growth, whereas a flat structure (basal root B6 in Fig. 1g) shows unidirectional root growth. This spatio-temporal 3D structure based analysis was used to explore the growth of basal (Fig. 2a; Video S5) and lateral roots of a bean seedling (Fig. 2b; Video S6), and lateral roots of a chickpea seedling (Fig. 2c; Video S7). All the basal roots emerged c. 12 h after transfer of the seedling to the growth system indicated by the emerging ridges. However, at 48 h the basal roots had different lengths. This analysis points to variations in growth rates as the sole responsible factor contributing to the differences in root length (Fig. 2d). These observations were confirmed in other plants of common bean (Fig. S6a,b). Although the lateral root initiation began at 30 h in the same plant (hidden under the basal roots in Fig. 2a), only three lateral roots emerged initially. However, at 48 h a large number of lateral roots began to emerge sequentially, with the lower ones emerging later (Fig. 2b,e). Other plants of common bean also showed similar behavior (Fig. S6c, Video S8). This initiation behavior was preserved even in secondary lateral roots when they grew from the basal roots (Fig. S6d, Video S8). A similar spatio-temporal pattern of lateral root initiation was also observed in chickpea, with a slight difference (Fig. 2c). The emergence of lateral roots in chickpea tended to have a rhythmic pattern. The uppermost lateral roots emerged within 0–16 h, followed by a delay of 8–16 h, after which a large number of lateral roots initiated within 32–48 h (Fig. 2c, f). Then, again, there was a pause of c. 12 h followed by initiation of the next batch of lateral roots. The rhythmic growth patterns of chickpea lateral roots are typical of plants of this variety (e.g. Fig. S6e,f).

image

Figure 2. Spatio-temporal structures constructed from segmented edge contours of (a) basal roots of a bean (Phaseolus vulgaris) seedling, (b) lateral roots of the same bean seedling and (c) lateral roots of a chickpea (Cicer arietinum) seedling. At the top of each spatio-temporal 3D structure, the last image of the sequence is shown to indicate the correspondence between the roots and the spatio-temporal 3D structures. Variations in length of selected roots from (a–c) are shown in (d–f), respectively, illustrating initiation and temporal changes in growth rate. The roots for which length vs time data are plotted in (d–f) are labeled in (a–c).

Download figure to PowerPoint

Analysis of local growth zone of roots

A second analysis targets local growth zones of individual roots. We identified the root tips from the edge contours, and determined the root midlines by passing a line through the maximum EDT from the contours of the root edges (Fig. S4) (Miller et al., 2007). Pixel intensities along the root midline at each time-step are presented as a function of space and time in Fig. 1(h). The striped patterns on each spatio-temporal intensity distribution indicate the movement of surface tissue. The region of the root where the stripes diverge with time shows the growth zone and the slopes of the stripes provide estimates of local growth velocities.

The root midline based analysis was tested in a basal (Fig. 3a, Video S2) and a lateral (Fig. 3b, Video S9) root of bean seedlings, and a lateral root of a chickpea seedling (Fig. 3c, Video S10). In all three cases, the patterns diverged with time. Treating the spatio-temporal patterns as streak lines of a 2D fluid flow, we calculated the spatio-temporal displacement vectors using the block matching technique of optical flow (Scarano, 2002; Basu et al., 2007) (Fig. S5a–c). From the displacement vectors the streak lines were obtained (Fig. S5d–f). The slope of the displacement vectors relative to the time axis is the root growth velocity, the spatial derivative of which is the REGR. As calculation of the derivative is highly affected by small variations, we interpolated the growth velocities along each streak line (to ensure that the velocities of the same tissue are used for interpolation) and smoothed before calculation of REGR. The calculations of REGR showed that in bean basal roots the unimodal growth zone bifurcated into a multimodal growth zone with two maxima as the peak splits into two ridges (Fig. 4a). In the lateral root of bean, after the 1-h mark, the growth zone shifted toward the apex (Fig. 4b), resulting in changes in the slopes of the patterns of graphite particles (black and white arrowheads in Fig. 3b). In the chickpea lateral root, the overall growth rate increased at the 2-h mark, which is indicated by the difference in the slopes of the black lines along the right edge of the REGR surface in Fig. 4(c). At this time the growth zone of the root also widened, implying that the increase in the growth rate of the root was associated with expansion of the growth zone rather than the local growth rate.

image

Figure 3. Analysis of the growth zone of the roots from the spatio-temporal patterns of graphite particles along the midlines of (a) a bean (Phaseolus vulgaris) basal root, (b) a bean lateral root and (c) a chickpea (Cicer arietinum) lateral root. The insets on the right of each panel show the root system and the white arrows point to the specific roots for which the data are presented. The insets at the top show close-ups of the corresponding roots with the midlines marked with white dashed lines. Therefore, the gray stripes in (a–c) are spatio-temporal patterns along these lines. The white arrowhead in (b) shows the spatio-temporal location where the initial basal growth zone ended and the black arrowhead shows the location from which a more apical growth zone evolved. The dark graphite stripes have a change in slope at these locations.

Download figure to PowerPoint

image

Figure 4. (a–c) Relative elemental growth rate (REGR) along the root midline is shown in space and time for the roots shown in Fig. 3(a–c), respectively. Both the height and shading of the spatio-temporal 3D plots show magnitudes of REGR.

Download figure to PowerPoint

The consequence of multimodal growth zones in the basal roots was observed in variations in the growth rate. As the basal root of bean in Figs 2(a), 3(a) was monitored for a longer time, we observed that these multiple local maxima of the growth zone grew and diminished transiently (Fig. 5a), which contributed to the rise and fall in growth rates of the basal roots (Fig. 5b). Such behavior of root growth was consistent in all the basal roots of bean (e.g. Fig. 5c,d).

image

Figure 5. (a) Relative elemental growth rate (REGR) of the bean (Phaseolus vulgaris) basal root of Fig. 3(a) is shown at a later stage. (b) Variations in the growth rate of the bean basal root of (a) are shown vs time. (c) An example of another bean basal root is shown with similar transient growth zones. (d) The growth rate of the bean basal root of (c) is shown vs time.

Download figure to PowerPoint

Root tip angle

The time-lapse movie (Video S2) showed that the basal roots had a wavy motion as they grew. Fig. 6(a) shows a superimposition of all root outlines of the bean seedling used in Fig. 1 (i.e. top view of Fig. 1(g) without the spatio-temporal surface). The root tips shown by the colored dots illustrate an oscillatory pattern, indicating wavy root growth – a reflection of the 2D projection of nutation of the roots (Video S2). By joining the root tips with the root bases (open circles in Fig. 6a), we calculated the tip angle θ of the root tips relative to the gravity. The tip angle changed with time as the roots followed a wavy growth pattern (Fig. 6b). Interestingly, the fluctuations in the root tip angle were greater in basal roots B1, B3 and B4, which were shorter in length than the other three basal roots. In addition, the roots tended to initially oscillate with a higher frequency (e.g. root B2) and with time the tip angles became relatively stable. Similar behavior was observed in all other bean seedlings.

image

Figure 6. (a) Superimposition of root outlines obtained from a time series of bean (Phaseolus vulgaris) basal root images. The colored dots show root tips and the open circles show root bases. (b) Root tip angles (θ) measured between the gravity vector and the lines joining root tip and base are shown as a function of time.

Download figure to PowerPoint

Assessment of the accuracy of the image analysis system

Analyses of the artificial root images for testing the accuracy of the image analysis system are presented in Figs 7, 8. Root edges were detected by livewire, from which spatio-temporal 3D structures were constructed. In Fig. 7(a), as the root elongated uniformly, the spatio-temporal 3D structure had a uniform slope at the root tip with time. The slope of the spatio-temporal 3D structure along the root tip in Fig. 7(b) changed, signifying changes in growth rate. Following identification of the root tips and the root midlines, root lengths were calculated and compared with the theoretical root lengths used in generating the artificial root images (Fig. 7c,d). The measured root lengths almost matched the theoretical root lengths (RMSD = 0.4 pixels).

image

Figure 7. (a–b) Spatio-temporal structures constructed from segmented edge contours of artificially generated root images. The red dots are the seed points, using which the edge contours (green lines) of the roots were generated. The blue spheres indicate the root tips and the magenta lines are root midlines. (c–d) Comparison of root lengths between theoretical values (solid lines) used in generating the images and calculated values (dashed lines) from the edge contours of the image sequences of (a) and (b), respectively.

Download figure to PowerPoint

image

Figure 8. (a–b) Spatio-temporal maps of midline intensities of the artificial root images. The black stripes in the background show the calculated spatio-temporal intensities of the marker dots and the overlying yellow lines show the theoretical spatio-temporal maps of the marker dots. (c–d) Theoretical relative elemental growth rate (REGR) distributions of the artificial root images used in generating the image sequences of (a) and (b), respectively. (e–f) Calculated REGR distributions corresponding to (c) and (d), respectively.

Download figure to PowerPoint

The spatio-temporal maps of midline intensities are shown in Fig. 8(a,b). The spatio-temporal traces of the marker dots (black stripes in the background) along the midline coincide with the theoretical spatio-temporal positions of the marker dots (overlaying yellow lines). The theoretical spatio-temporal variations of REGR used in generating the sequences of images with single and bifurcating growth zones are shown in Fig. 8(c,d), respectively, and the corresponding REGR distributions calculated from the images are shown in Fig. 8(e,f). The RMSD values are 0.0034 per time-point (between Fig. 8c and Fig. 8e) and 0.0062 per time point (between Fig. 8d and Fig. 8f). With respect to the peak REGR value (0.05 per time-point), differences between the theoretical REGR and the calculated REGR are 6.7% and 12.5% for single and bifurcating growth zones, respectively.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We present here a new semiautomatic methodology to visualize and quantify initiation and growth of roots in both space and time. The methodology addresses both overall growth and local growth zones of roots, providing interesting insights into the biology of root system development. The methodology is tested in basal and lateral roots of bean seedlings and lateral roots of chickpea seedlings. The test results demonstrate that both overall growth and local growth zones are analyzable using our methodology, which has the potential to reveal hitherto unexplored patterns of root development.

The new methodology begins with the segmentation of the roots from the background, which is typically an arduous task. Therefore, we developed a novel technique where the user input is required in only a few sets of images to generate interpolated images on arbitrary slicing planes followed by livewire-assisted semiautomatic segmentation of the roots on these spatio-temporal slicing planes. Once this step has been completed, the rest of the process is nearly automated. The number of slicing planes used for generating spatio-temporal interpolated images is not dependent on the number of original images. Therefore, even for a large number of images, the human input does not need to increase, although the computational burden increases proportionately. Once the root edge contours have been detected, further analyses are performed to obtain information relevant for exploring the development of root system architecture. For a set of 30 images of 10 megapixel resolution it takes c. 1 h to do a complete analysis on a 2-GHz Intel® Core 2 Duo computer with 2 GB RAM. This includes time for both computer processing and user input.

The formation of spatio-temporal 3D structures from root edge contours is a new step that this technology introduces for exploring the initiation of secondary roots such as basal and lateral roots. As evident in the test cases (Fig. 2), it is very difficult to exactly pinpoint when a basal or a lateral root emerges as the process is continuous. In addition, measurement of structural details, for example, the length and angle of a barely emerging root, can be erroneous as both the root base and the tip are not uniquely identifiable even from the high-resolution images. Therefore, instead of direct quantitative assessment of root emergence, a somewhat qualitative description is preferred. The spatio-temporal 3D structures provide such information, as can be seen in the emergence of the ridges in Fig. 2(a–c). This 3D structure can help identify the spatio-temporal window when the secondary roots begin to emerge, rather than being over-specific. Thus, the spatio-temporal 3D structure allows visualization and assessment of root initiation, branching and spatio-temporal changes in length, diameter and angle. Furthermore, such analysis shows the development of multiple roots simultaneously, allowing comparative studies of the architectural behaviors. In the test cases here, we demonstrated that the heterogeneity of lengths of bean basal roots arose from differences in growth rates only as the emergence of basal roots was nearly simultaneous. However, for the lateral roots of both bean and chickpea, root lengths depended on both emergence time and growth rate. We also found that chickpea lateral roots grew with a specific rhythm, similar to a phenomenon reported in Arabidopsis thaliana (Lucas et al., 2008). At each time window a group of lateral roots emerged, followed by a pause, and then the next set emerged. However, bean lateral roots emerged sequentially without any such pause or rhythm. Therefore, the current technology paves the way for detailed investigations of how such growth patterns help these different plants adapt to specific environmental conditions.

The local growth analysis using spatio-temporal patterns of pixel intensities along the root midline allows a visual demonstration of the spatio-temporal growth zones of the roots and, consequently, helps to elucidate the dynamics of root development in both space and time in finer detail. The measurements from such an approach show that there are variations of REGR in bean and chickpea lateral roots, indicating the transient nature of the growth zones, which could have been missed without the REGR distribution in the space–time continuum. Interestingly, bean basal roots show a clearly bifurcating multimodal transient growth zone, which grows and diminishes. Similar transient growth zones were also observed in maize (Zea mays; Walter et al., 2002).

The lateral movement of the root tip was visualized when the root outlines and tips from all images were superimposed. We found a wavy motion of the root tip, which was higher in smaller roots. The root tip angle was calculated by θ = tan−1 (a/b), where a and b are the horizontal and vertical distances of the root tip from the base, respectively. For very small roots, even a small growth can change the ratio a/b very rapidly, resulting in rapid fluctuations in root tip angle. Therefore, although root initiation angle has been identified as an important determinant of spatial localization of roots (Clark et al., 2011), one has to be careful in quantifying the initiation angle because of rapid temporal variation at the initial stages of seedling development.

The methodology presented here was tested by analyzing artificial root images for which the growth parameters were known. The results showed highly accurate assessment of the root lengths, indicating precise edge detection by livewire. Although the diverging patterns of spatio-temporal maps of root midline intensities indicate approximate growth zones, without further quantitative analysis it is impossible to estimate the nature of the growth zones. For example, only visual examination of Fig. 8(a,b) does not reveal that in one case (Fig. 8a) there is a unimodal growth zone, whereas in the other (Fig. 8b) the growth zone bifurcates. Thus, the new technology not only provides a visual demonstration of the growth zones through spatio-temporal maps of root midline intensities but also provides tools to quantify the local growth zones. Although the comparison of REGR between theoretical and calculated values indicated qualitative similarity, the %RMSD values were slightly higher. Further analysis indicated that the higher values of %RMSD were contributed by three sources. First, the coordinates of the marker dots calculated for each image were typically decimal point numbers, but the images required these coordinates to be integers, resulting in rounding errors. With higher resolution images the rounding errors can be reduced, but at the cost of computing efficiency. Secondly, calculation of REGR required calculation of the derivative of root growth velocity which, in turn, required smoothing of the root growth velocity. A higher amount of smoothing caused smoother REGR distributions, but also underestimated the peak REGR values as smoothing reduced the slopes of root growth velocity, and vice versa. Because the growth velocities were smoothed along the streak lines, REGR distributions along the streak lines appeared smoother and sharper (e.g. the bifurcated REGR peak toward the tip in Fig. 8f). Finally, the calculation of root growth velocity and REGR depended on pattern matching, and therefore, on the patterns available. A higher number of marker dots of distinct patterns aided estimation of growth velocity and REGR.

Although we used the methodology to analyze root growth, it can be used for analyzing growth of any tissue. There is also no specific requirement of image source for use of the technique. However, similar to any other image analysis technique, the quality and reliability of assessments from the images depend on the clarity of the images. The images used to test the system were of high quality and required little human intervention. However, on rare occasions when the images were of poorer quality because of condensation on the glass in front of the roots or uneven illumination, the automated segmentation failed. In those cases the user had to manually segment the roots. As the methodology uses 2D images, it requires that the growth be uniplanar. In the current study this was ensured by the transparent growth system, which may not be the case for other tissues. It has been shown previously that the root length and angle are directly correlated between 2D culture and sand or soil culture (Liao et al., 2001). Optical flow based analyses for assessment of velocity are sensitive to image quality and parameters governing the calculations. But, in this method, the visual demonstration of pattern movement before optical flow analysis provides an opportunity to verify the calculated velocities and streak lines against the patterns, and correspondingly adjust the optical flow parameters for accurate results. Therefore, the new methodology offers the possibility of unraveling newer and greater details of developmental and growth patterns and, hence, may prove to be very useful in investigations of tissue growth in biological systems in general. The image analysis software is available from the website http://home.iitk.ac.in/~apal/growthexplorer.html.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We thank Dr Partha S. Basu at the Indian Institute of Pulses Research, Kanpur, India for providing the legume seeds. This work was financially supported by a grant from the Fast Track scheme by Department of Science and Technology, Government of India (no. SR/FT/LS-085/2007).

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Fig. S1 Flowchart of the image analysis system.

Fig. S2 Illustration of the vibration stabilization process.

Fig. S3 Edge detection by livewire algorithm.

Fig. S4 Identification of the root midline using the Euclidean distance transform.

Fig. S5 Root growth velocity and growth trajectories.

Fig. S6 Examples of spatio-temporal 3D reconstructions of basal and lateral roots.

Notes S1 Further details of vibration stabilization and image analysis.

Video S1 Time-lapse movie of a growing bean seedling before stabilization showing the effect of vibration caused by the ventilation fans in the growth chamber; after stabilization, this movie is shown as Video S2.

Video S2 Stabilized time-lapse movie of a growing bean seedling showing the development of basal roots during a period of 5 h; dark spots are graphite particles sprinkled on the roots for analysis of local growth.

Video S3 Movie showing growth of an artificially generated root with a unimodal growth zone; black dots were used as marker dots to assess local growth zones.

Video S4 Movie showing growth of an artificially generated root with bifurcating growth zones; black dots were used as marker dots to assess local growth zones.

Video S5 Time-lapse movie of a bean seedling showing emergence and development of basal roots for 48 h.

Video S6 Time-lapse movie of the same bean seedling in Video S2 for 48–70 h showing emergence and growth of lateral roots.

Video S7 Time-lapse movie showing initiation and development of lateral roots of a chickpea seedling during 72 h.

Video S8 Time-lapse movie a bean seedling for 124 h showing emergence and growth of basal and lateral roots; the faint white rectangles show the segments in which spatio-temporal growth patterns of lateral roots were studied in Fig. S6(c,d).

Video S9 Time-lapse movie showing local growth of bean lateral roots during a period of 5 h following sprinkling of graphite particles.

Video S10 Time-lapse movie showing local growth of chickpea lateral roots during a period of 5 h following sprinkling of graphite particles.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

FilenameFormatSizeDescription
NPH_4149_sm_FigS1-S6-NotesS1.doc5321KSupporting info item
NPH_4149_sm_VideoS1.mov1340KSupporting info item
NPH_4149_sm_VideoS2.mov641KSupporting info item
NPH_4149_sm_VideoS3.mov61KSupporting info item
NPH_4149_sm_VideoS4.mov61KSupporting info item
NPH_4149_sm_VideoS5.mov2143KSupporting info item
NPH_4149_sm_VideoS6.mov924KSupporting info item
NPH_4149_sm_VideoS7.mov1158KSupporting info item
NPH_4149_sm_VideoS8.mov566KSupporting info item
NPH_4149_sm_VideoS9.mov829KSupporting info item
NPH_4149_sm_VideoS10.mov347KSupporting info item