EZ-Rhizo: integrated software for the fast and accurate measurement of root system architecture

Authors


(fax +44(0)1413304447; e-mail P.Armengaud@bio.gla.ac.uk).

Summary

The root system is essential for the growth and development of plants. In addition to anchoring the plant in the ground, it is the site of uptake of water and minerals from the soil. Plant root systems show an astonishing plasticity in their architecture, which allows for optimal exploitation of diverse soil structures and conditions. The signalling pathways that enable plants to sense and respond to changes in soil conditions, in particular nutrient supply, are a topic of intensive research, and root system architecture (RSA) is an important and obvious phenotypic output. At present, the quantitative description of RSA is labour intensive and time consuming, even using the currently available software, and the lack of a fast RSA measuring tool hampers forward and quantitative genetics studies. Here, we describe EZ-Rhizo: a Windows-integrated and semi-automated computer program designed to detect and quantify multiple RSA parameters from plants growing on a solid support medium. The method is non-invasive, enabling the user to follow RSA development over time. We have successfully applied EZ-Rhizo to evaluate natural variation in RSA across 23 Arabidopsis thaliana accessions, and have identified new RSA determinants as a basis for future quantitative trait locus (QTL) analysis.

Introduction

Uptake of water and mineral nutrients from the soil is essential for plant life. The plant root system achieves these fundamental functions through its highly responsive and plastic morphology, which allows the plant to adjust to and exploit the wide spectrum of physical and chemical soil properties. Root system architecture (RSA), the geometry of different parts of the root system, is the overall output of growth and development of individual cells and tissues within the root. RSA feeds back into plant growth and development, at the very least by determining the rate of water and nutrient uptake. Furthermore, it acts as a ‘sensory’ system, because it exposes different parts of the roots to different soil microenvironments. RSA is therefore an important trait defining agricultural productivity, and represents a model system for fundamental research in plant development and signalling.

The RSA is broadly determined by the genetic make-up of the plant, and is subject to the abiotic and biotic environment of the root as well as intrinsic factors related to the physiological status of the plant (Malamy, 2005). The availability of nutrients, such as nitrogen (N), phosphorus (P) and potassium (K), best illustrates the influence of the soil environment on RSA. Plants grown in low-N or -P regimes display relatively longer lateral roots than plants grown in high-N or -P regimes (Linkohr et al., 2002; Stitt and Feil, 1999; Zhang et al., 1999), whereas deficiency in K reduces lateral root length (Armengaud et al., 2004). Furthermore, the RSA varies not only according to environmental nutrient concentration, but also with the internal nutrient resources of the plant (Williamson et al., 2001). Such behaviour implies that nutrient sensing and signalling affect root-specific morphologic responses, which in turn can be used to identify the underlying molecular components. Several genes have been implicated in the physiological responses of plants to P deficiency (e.g. PHO2, PHR, mi399 and LPR1; Bari et al., 2006; Svistoonoff et al., 2007), and may also play a role in determining RSA under P deficiency. However, doubt has recently been cast on the role of P deficiency per se in regulating RSA, the suggestion being that the reported responses are caused by the Fe toxicity that occurs in low-P medium (Ward et al., 2008). Some genes with specific functions in the signalling, transport and metabolism of N have been found to modulate RSA responses (e.g. ANR1, NRT1.1 and NRT2.1; Little et al., 2005; Remans et al., 2006; Zhang and Forde, 1998). Nevertheless, most of the genes involved in the control of RSA remain unknown.

Cell-cycle regulators and hormones (especially auxin) have been described as regulators of root growth parameters, notably in the dominance of the primary root tip, lateral root initiation and growth. The stimulatory effect of auxin on lateral root development is well known, and is commonly exploited in horticulture (Casimiro et al., 2003). Furthermore, auxin signalling and transport mutants show altered gravitropic responses, which can be quantified by determining vertical and horizontal growth indices as measures of root geometry (Grabov et al., 2005). Cytokinins, ethylene, abscisic acid and brassinosteroids also influence RSA, by generally decreasing root size and delaying development (Malamy, 2005; Osmont et al., 2007).

The RSA is a quantitative trait, and genetic approaches have been used to identify genes that mediate between intrinsic or environmental factors and RSA. Quantitative trait loci (QTLs) were identified based on the natural variation of Arabidopsis thaliana under standard conditions (Loudet et al., 2005), and in response to osmotic stress (Fitz Gerald et al., 2006), nutrient deprivation (Rauh et al., 2002; Reymond et al., 2006; Svistoonoff et al., 2007) or toxicity (Hoekenga et al., 2003; Kobayashi and Koyama, 2002; Kobayashi et al., 2007). Such approaches were particularly successful using Bay0 × Schakdara recombinant inbred lines to identify genes involved in the sensing of P and sulphur (S) (Loudet et al., 2007; Svistoonoff et al., 2007).

Only a small number of quantitative studies have taken account of RSA parameters other than the main root length. One of the few examples is the analysis of main root length, lateral root number and lateral root density in a Bay0 × Schakdara population, which clearly showed that consideration of these additional RSA parameters uncovers novel QTLs (Loudet et al., 2005). Considering that RSA is a complex trait, it is unlikely that considerable progress in identifying the underlying processes can be made without efficient tools to capture the RSA comprehensively.

Information is also sparse on how RSA responds to combined stimuli: e.g. different ratios of nutrients, varying carbon/nitrogen status, and combined nutritional and abiotic stresses. Again, to remedy this knowledge gap we require tools to quantify RSA efficiently from large numbers of plants. Current approaches use rulers or digital callipers to determine the main root length, or use commercial software such as ImageJ (Abramoff et al., 2004), Optimas analysis software (Media Cybernetics, http://www.mediacy.com) or WinRhizo (Arsenault et al., 1995) to assess lateral root length and number. The software programs provide accurate and reliable results, but require considerable time (for arranging the harvested plant material and specifying different parts of the root), and are therefore unsuitable for a comprehensive analysis of RSA. In this paper, we describe a reliable, semi-automated and easy-to-use software application, EZ-Rhizo, which was developed in our laboratory for measuring multiple RSA parameters of young plants grown in vertical Petri dishes. We illustrate the performance of EZ-Rhizo by describing the natural variation of root architecture in 23 accessions of A. thaliana.

Results

Quantitative analysis of RSA using EZ-Rhizo

We have divided the process of root architecture analysis between three basic work steps: image acquisition, RSA measurement and data analysis. These steps are independent from each other, and can be carried out at different times.

Image acquisition.  EZ-Rhizo analyses 2D images of plant roots obtained from plants growing on solid support, such as A. thaliana plants growing on agar in vertical Petri dishes. Plants are scanned at chosen time intervals, and the images are saved in bitmap format. A scanning resolution of 200 dots per inch (dpi) was found to be the best compromise between image quality and analysis time. It allows detection of lateral roots of at least 0.1 mm in length. Image acquisition is non-invasive, and hence the growth of individual plants can be monitored over time.

RSA measurement.  The measurement of the RSA is carried out with the EZ-Rhizo program. After opening an image file, the user supervises the RSA measurement by performing a number of pre-defined operations (offered as toolbar icons in the EZ-Rhizo menu, see Figure 1a). The first four operations prepare the image for subsequent root detection.

Figure 1.

 Screenshots of the EZ-Rhizo program and its raw data output.
(a) Screenshot of the EZ-Rhizo main window. The original image shown on the left was taken through the standard set of pre-defined operations (performed with toolbar icons), leading to automatic root detection by the program (right-hand side) and measurement of root system architecture (RSA) parameters (see Table 1), which were saved as a text file.
(b) Screenshot of part of the data output in the text file. The data was subsequently integrated into a searchable SQL database (last two icons in the toolbar shown in A).

  • 1 ‘Make black and white’: any colour image is converted into a binary image. The user chooses the threshold value on a 255-RGB scale (the default is 90). The modified image is immediately displayed in the image preview window, as it is throughout all subsequent steps.
  • 2 ‘Remove box’: this step removes any outlying edges that appear in the image, such as the ones associated with the rim of an agar plate. Upon a mouse click, the image is automatically cropped.
  • 3 ‘Remove noise’: a common problem associated with defining image masks is eliminating background noise. EZ-Rhizo offers a selection of erosion methods (Gaussian, Gaussian-2, exponential, pyramid shaped, average and conical) from which to choose. The default is Gaussian-2, but optimization of the signal/noise ratio applying other options is performed in consultation with the image preview.
  • 4 ‘Dilate’: a pixel dilation function is automatically applied to fill any gaps within the roots created by the removal of background noise. The following operations are used to identify and then numerically analyse RSA.
  • 5 Skeletonize: this operation removes pixels from the edges of the mask objects to give single pixel-wide skeletons using a non-interrupted skeletonization algorithm developed by Saeed et al. (2001). Skeletonization is essential, as the following analysis algorithms for root-path tracking and branch detection work with single pixel-width objects.
  • 6 ‘Re-touch’: this function allows manual image editing in a minimal length of time. The user can apply a white brush for re-connecting any discontinuities in the root path created by previous steps, and a black brush for separating roots from shoots.
  • 7 ‘Find roots’: roots are detected using a purpose-built, image-scanning algorithm that identifies all objects formed from contiguous white pixels (Freeman, 1974). After completion, every detected object is displayed in the image window (Figure 1a).
  • 8 ‘Confirm roots’: dialogue boxes present each detected object to the user to be confirmed or rejected as a root with a single mouse click.
  • 9 ‘Save experiment’: based on an algorithm developed by Kimura et al. (1999), RSA parameters are automatically measured from the pixel tree structure of each confirmed object. At this stage, the user is prompted for the relevant metadata of the image (e.g. user, experiment name, medium and resolution), before saving all measured RSA parameters in a text file. A screenshot of a results file in text format is shown in Figure 1b. All RSA parameters determined by EZ-Rhizo (measured and derived) are listed and defined in Table 1.
Table 1.   Measured and derived root system architecture (RSA) parameters determined with EZ-Rhizo
 Measured parametersDerived parametersDefinition
  1. aVector: shortest line connecting origin of main root (or lateral root on main root, or lateral rootn + 1 on lateral rootn) with tip of main root (or lateral root, or lateral rootn).

  2. bWith a resolution of 0.01 cm for a 200-dpi image.

Main root (MR)Length Path length of MR (cm)
Vector length Length of MR vectora (cm)
Angle Angle between vector and absolute verticality (°)
Number of LR Number of LR detected on MRb
 StraightnessVector length divided by path length
 DepthLength of vector projection onto a vertical line through MR origin (cm)
 Basal ZoneLength of path from origin of MR to first LR (cm)
 Branched Zone (BZ)Length of path from first (top) to last (bottom) LR (cm)
 Apical ZoneLength of path from last lateral root to tip of MR (cm)
 LR density/MRNumber of LR divided by MR path length (cm−1)
 LR density/BZNumber of LR divided by BZ (cm−1)
Lateral root (LR)Position Length of path from origin of MR to origin of LR (cm)
Length LR path length (cm)
Vector length Length of LR vector (cm)
Angle Angle between vector and absolute verticality (°)
Number of LR2 Number of 2nd order LR on LR
 StraightnessVector length divided by path length
Higher order lateral roots (LRn)Order Order n of LRn (n > 1)
Position Position on LRn – 1. Length of path from origin of LRn – 1 to origin of LRn (cm)
Length LRn path length (cm)
Vector length Length of LRn vector (cm)
Angle Angle between vector and absolute verticality (°)
Number of LRn + 1 Number of next higher order LR on LRn
 StraightnessVector length divided by path length
Total Root System (TRS) Total root system sizeSum over path lengths of MR and all LR (cm)

Data analysis.  EZ-Rhizo can retrieve data from the results text files and add them into a database (using the open-source MySQL application, which is provided as an optional download during EZ-Rhizo installation). The program subsequently allows the user to query the MySQL database for measured and derived RSA parameters (see Table 1), as well as for metadata (e.g. date and media), and then to save the selected data in comma-separated value (csv) file format for further analysis (e.g. in Excel). Scripts and instructions (available at http://www.ez-rhizo.psrg.org.uk) have also been prepared to directly access the database, and to perform statistical analyses using the ‘R’ package (R Development Core Team, 2003).

Natural RSA variation across A. thaliana accessions

We have used the EZ-Rhizo software to evaluate the RSA variation across 23 A. thaliana accessions (for information on individual accession see http://www.arabidopsis.org or http://dbsgap.versailles.inra.fr/vnat/). Exemplary root images of some accessions and their detection by EZ-Rhizo are shown in Figure 2. Seeds were collected from plants grown under identical conditions: surface sterilized and germinated on half-strength MS medium (Murashige and Skoog, 1962) in vertical Petri dishes. The RSA was measured every other day from 3 to 9 days after germination. At the last stage, the total root system size was calculated by adding the lengths of all of the lateral roots to the length of the main root. Figure 3 shows the range of total root system size displayed by the accessions at day 9. The average total root system size was 7.9 cm, and showed a variation of 39% across accessions. WS, Lip0 and Col0 presented the biggest root systems (17.0, 13.6 and 12.6 cm, respectively), whereas Bu2 and Ei2 had the smallest roots, of approximately half the size of the population average.

Figure 2.

 Natural variation of root system architecture in Arabidopsis thaliana accessions.
Typical root architectures of A. thaliana accessions Bu2, Bla11, Wei1, Col0, Ler0 and WS. For each root the original image is shown on the left; the object detected and measured by EZ-Rhizo is shown on the right.

Figure 3.

 Variation of total root system size across 23 Arabidopsis thaliana accessions.
Plants were grown for 9 days after germination on half-strength MS medium. The total root system size was calculated by summing path lengths of the main root and all lateral roots of individual plants. Means (±SE, n = 8 plants) for each accession are sorted in descending order.

Using the measured positions of the first and last lateral roots we subdivided the main root into basal, branched and apical zones (Table 1). Accessions differed in the relative contribution of these zones to the total main root length (Figure 4). Generally, accessions with longer main roots displayed a ratio of apical zone to branched zone of around 1, whereas in accessions with shorter main roots this ratio was larger. Bla11 and Pyl1 differed markedly from the other accessions: the former had a relatively long basal zone, whereas the latter displayed no branched zone.

Figure 4.

 Variation of root zoning across 23 Arabidopsis thaliana accessions.
Path lengths of basal (white), branched (grey) and apical (black) root zones (for definitions, see Table 1) were determined for plants grown for 9 days after germination on half-strength MS medium. Means (±SE, n = 8 plants) for each accession are sorted by descending main root (MR) path length.

Principal components determining natural RSA variation

To identify the determinants of RSA variability across A. thaliana accessions, we applied principal component analysis (PCA; Jolliffe, 2002) to the full data set obtained with EZ-Rhizo (Table S1). Prior to PCA, a Pearson correlation matrix was established to identify and eliminate redundant parameters. Table 2 summarizes the degree of correlation among a selected subset of 10 parameters that were subsequently used for PCA. PCA uncovered that 88% of the variation in RSA is captured by four principal components (PC1–PC4; Table 3). However, eight principal components are needed to cover the full variation. This finding illustrates that RSA variability cannot be fully represented if only a few RSA parameters are measured. The first component represents 39% of the variability, and consists mostly of lateral root number (Table 3); 22% of additional variation projected by PC2 comes primarily from the main root length. PC3, PC4 and PC5 add, respectively, another 17, 10 and 6% of the observed RSA variability: these components account mainly for lateral root density within the branched zone, main root angle and average lateral root length (Table 3). Plots of accession distribution for different combinations of the first four principal components are presented in Figure 5.

Table 2.   Pearson correlation matrix of root system architecture (RSA) parameters
RSA parametersMR lengthMR angleLR no.LR lengthLR angleLR density/MRBasal zoneApical zoneBZLR density/BZ
  1. Bold values show significance at P < 0.01.

MR length10.0430.5730.2570.0250.1760.1550.8140.724−0.025
MR angle0.0431−0.116−0.0160.007−0.1280.1080.137−0.0860.080
LR number0.573−0.11610.599−0.1870.878−0.3800.0980.948−0.357
LR length0.257−0.0160.5991−0.5640.571−0.4270.0020.572−0.174
LR angle0.0250.007−0.187−0.5641−0.2980.631−0.037−0.057−0.310
LR density/MR0.176−0.1280.8780.571−0.2981−0.529−0.3060.725−0.302
Basal zone0.1550.108−0.380−0.4270.631−0.52910.172−0.209−0.125
Apical zone0.8140.1370.0980.002−0.037−0.3060.17210.2510.233
BZ0.724−0.0860.9480.572−0.0570.725−0.2090.2511−0.419
LR density/BZ−0.0250.080−0.357−0.174−0.310−0.302−0.1250.233−0.4191
Table 3.   Variable square cosines of root system architecture (RSA) principal component analysis
 PC1PC2PC3PC4PC5PC6PC7PC8
  1. Bold values show the highest contribution to the component.

MR length0.2930.6360.0560.0090.0010.0010.0010.001
MR angle0.0150.0310.0530.8830.0170.0010.0000.000
LR no.0.9230.0110.0220.0000.0320.0010.0020.000
LR length0.5830.0460.0360.0210.2070.0480.0580.001
LR angle0.1500.2460.4480.0000.0640.0010.0910.002
LR density over MR0.7370.0950.0480.0030.0700.0170.0130.014
Basal zone0.2490.3940.1140.0030.0480.1580.0320.000
Apical zone0.0140.6130.2950.0130.0060.0490.0000.010
Branched zone (BZ)0.8400.1080.0290.0000.0050.0000.0010.017
LR density over BZ0.1030.0070.6020.0300.1310.1170.0080.000
Variability (%)39.121.917.09.65.83.92.10.4
Cumulative variability (%)39.160.978.087.693.497.399.499.8
Figure 5.

 Principal component analysis of natural root system architecture (RSA) variation in Arabidopsis thaliana.
A total of 10 RSA parameters were used to analyse the natural variation across 23 accessions (see Table 3). The position of accessions in 2D plots are shown for principal component (PC) 2 against PC1, representing 61% of the variability (A); PC1 and PC3 against PC1 (56%; B); PC4 against PC1 (49%; C); PC5 against PC1 (45%; D); and PC5 against PC2 (28%; E).

The plot of PC2 against PC1 (Figure 5a) positions WS, Lip0 and Col0 at the extreme right, as they have abundant lateral roots, and Bu2, Ei2 and Pyl1 at the extreme left, with fewest lateral roots. PC2 separates Da0 (long main root) from Sorbo (short main root). Mixed properties were found for Kl0, Ru1 and Tsu1 on the one hand, and for Schakdara and Kondara on the other hand. The first group displays relatively long main roots with relatively few lateral roots; the second group showed the opposite features. Bla11 appeared as an outlier with a very long main root and very few lateral roots.

Figure 5b shows how the accessions vary with respect to lateral root number (PC1) and lateral root density within the branched zone (PC3). The two components account together for 56% of RSA variability. However, this value is strongly influenced by Pyl1, which has a very low lateral root number (0.63 ± 0.3 per main root on average) resulting in a very small branched zone, and an artificially high lateral root density. The 2D presentation separates Schakdara (high number and high density of lateral roots) from an accession cluster including Bla11, Ei2, Bu2 and Fei0 (low number and low density of lateral roots) on the positive diagonal, and Wei1 (high number and low density of lateral roots) from Tsu1 and Cvi0 on the negative diagonal (low number and high density of lateral roots).

Figure 5c illustrates that 49% of RSA variability is based on the lateral root number (PC1) and main root angle (PC4). Identification of the latter as an important determinant of RSA variability was unexpected. This parameter, which was consistent throughout growth from day 3 to day 9, may reflect the variability of gravitropic control or rotational rhythms. The main roots of accessions Da0, RRS7, Col0, Ang0, Ms0 and Bu2 grew almost vertically (main root angle <2°). By contrast, in Cvi0 and Ler1 the main root angle was much higher (12.9° and 13.1° at day 3), and increased further with time (20.4° and 18.7° at day 9).

The variability in the average length of lateral roots is represented by PC5, and accounts for 6% of RSA variability. Figure 5d shows how A. thaliana accessions were positioned with respect to PC1 and PC5 (45% of variability). For example, Schakdara and Mh1 are similar with respect to the number of lateral roots, but differ strongly in the average lateral root length.

Figure 5e shows the arrangement of accessions with respect to PC2 and PC5. The overall contribution of these two components to RSA variability is relatively low (28%), but they are interesting as they illustrate the relative growth investment into the main root (PC2) and the lateral roots (PC5). For example, Bla11 has a long main root (7 cm on average) but relatively short lateral roots (0.1 cm). Sorbo shows the opposite pattern, with long lateral roots (0.4 cm) and a short main root (3.9 cm). Schakdara presents the second longest average length of lateral root (0.5 cm) after WS, and a main root length of just about the population average (5.6 cm).

In summary, the PCA analysis of EZ-Rhizo data reveals the relative importance of individual RSA parameters for genetic variation, and highlights accession pairs that could be crossed to identify the genetic basis of specific RSA traits (e.g. Pyl1 × Lip0 for the lateral root number trait; Da0 × Sorbo for the main root length trait).

Natural variation in lateral root patterns

The average lateral root length is a significant contributor to RSA variability (Figure 5d,e; Table 3), but it is a complex feature as plants have both long and short lateral roots, depending on their date of emergence and growth rate. We analysed the pattern of lateral root growth in more detail by taking into account the position of individual lateral roots on the main root. Lateral root length averages were calculated separately for six root segments comprising the basal 60% of the main root path (no lateral roots were observed beyond this point). Three representative patterns are shown in Figure 6 (see Figure S1 for the patterns of all accessions). In most accessions the basal lateral roots are longer than the apical ones, creating a typical inverted pyramid pattern. Accessions with high average lateral root length generally show a very steep increase in lateral root length towards the root base (Figure 6a). Among the accessions with lower average root length there is considerable variation in lateral root patterns, some showing a steep slope of lateral root length along the main root axis (Figure 6b), and others displaying a much shallower pattern (Figure 6c). It is clear from this basic analysis that by measuring the position of individual lateral roots, EZ-Rhizo can provide an important numerical input into spatial models of lateral root growth, and can thereby identify novel phenotypes.

Figure 6.

 Spatial patterns of lateral root length.
Mean lateral root lengths (±SE) over all lateral roots detected within a given root segment for a selection of Arabidopsis thaliana accessions. Root segments are given as a percentage of the main root path starting from the root base (top). Three typical patterns (A–C) are shown, and accessions that showed similar patterns are listed on the right. All patterns are shown in the Figure S1. In total, 1179 lateral roots from the 23 9-day-old accessions were analysed.

Natural variation of root growth

Image acquisition for EZ-Rhizo is non-invasive, and data created with EZ-Rhizo can therefore be used to gain information on the dynamics of RSA. For this study we obtained images at 3, 5, 7 and 9 days after germination, but the simplicity of image processing by EZ-Rhizo means that a much higher time resolution of RSA can be achieved in future studies. In Figure 7a, accessions are sorted by the average growth rate of the total root system (TRS) over the entire growth period between days 3 and 9. Figure 7a also shows the contribution of the main root growth rate, which indicates how the total root growth is divided between the main and lateral root types. Within the developmental window assessed here, WS, Lip0 and Col0, the fastest root growers, invest almost equally into the growth of main and lateral roots, whereas other accessions (e.g. Pyl1, Bla11, Tsu1 and Da0) strongly favour growth of the main root.

Figure 7.

 Variation of growth rate and growth acceleration of the total root system (TRS) and the main root (MR).
(a) Average growth rates over the entire period of time between 3 and 9 days after germination were determined for individual plants. (b) The slope of a linear regression fit to three growth rates (days 3–5, 5–7 and 7–9) was calculated for individual plants (growth acceleration). Histograms show the means ± SE (n = 8 plants) for each accession.

In most accessions, growth rates were not constant over time, but rather increased during early plant development. To quantify this growth acceleration, for each plant we determined the slope of a linear regression performed on the growth rates measured between days 3 and 5, 5 and 7, and 7 and 9 (Figure 7b). This operation increased the resolution for finding differences in root growth between the accessions. Thus, variation of TRS growth rate among accessions is 40%, whereas variation of growth acceleration is 58%. Variation of the respective parameters for the main root is smaller (22 and 24%), indicating that acceleration of growth in the total root system is mainly determined by accelerated lateral root growth (at least over the time period assessed here). The lateral root growth rate is again a complex parameter, as individual lateral roots show different growth rates depending on their position on the main root. When the lateral root growth rate was calculated for lateral roots located within the main root sections between days 7 and 9, we observed much higher values for the basal lateral roots than for the apical ones (Figure S2). Among the 23 accessions examined, WS, Col0 and Schakdara show the highest basal lateral root growth rate (around 2 mm day−1), and Fei0, Kl0 and Wei1 show the lowest rate (around 0.5 mm day−1). Natural variation is also evident for the difference between basal and apical lateral root growth. For example, in Ler1 and Ang0 basal lateral roots grew approximately three times faster than apical ones, whereas in Bla11, Ct1 and WS the difference of growth rate between apical and basal lateral roots was more than 10-fold. Such detailed data clearly show that EZ-Rhizo can make an important contribution to the kinetic analysis and modelling of root growth in different accessions and mutants.

Discussion

Features of EZ-Rhizo

We have presented a new method for the quantitative characterization of the RSA from plants grown against a solid substrate, and have illustrated its application with A. thaliana accessions grown in vertical Petri dishes. The EZ-Rhizo method combines non-invasive image acquisition with a newly developed software for root detection, measurement of multiple RSA parameters, data storage and data analysis. The following features of EZ-Rhizo mark a considerable improvement over previously available methods.

  • (i) Speed: getting from the digital image to a full set of RSA parameters for all plants on the plate takes only a few minutes.
  • (ii) Semi-automation: whereas a certain degree of user supervision is required for image preparation, the root detection and measurement of all RSA parameters are automated. Automation not only increases the speed of the analysis, but also eradicates bias introduced by the researcher.
  • (iii) Ease of use: user supervision of the measurement is carried out by simple, pre-defined cursor operations, the results of which are immediately displayed in the image window.
  • (iv) Accuracy: a pixel-based tree algorithm allows the detection of root paths with much higher accuracy than can be achieved using digital drawing tools.
  • (v) Non-invasiveness: the method is non-invasive, and thereby allows the user to monitor individual RSA parameters over time.
  • (vi) Flexibility in data management and mining: the data input and output uses commonly used file formats (.bmp, .txt and. csv), which are suitable for subsequent data analysis with different third-party programs. The optional data storage in the MySQL database allows for the customized querying of the data set.
  • (vii) Integration with modelling tools: data created with EZ-Rhizo can easily be integrated into programs simulating root growth and development.
  • (viii) Availability: EZ-Rhizo is freely available to the academic science community through our EZ-Rhizo website, http://www.ez-rhizo.psrg.org.uk which also provides user support.

Potential problems: solutions and limitations

The analysis of RSA from plants grown on vertical dishes can pose a number of problems caused either by the growth system or by the root system itself. In the following text we outline whether and how EZ-Rhizo addresses these problems.

Image quality.  Plant growth on agar plates presents a number of challenges to root detection and measurement, including, for example, interference from condensation droplets, and low contrast between roots and media. Although these problems can be minimized experimentally (e.g. using appropriate temperatures and dyes), they cannot be avoided altogether. EZ-Rhizo is effective in removing background noise by applying a variety of filters (‘remove noise’ function), and artefacts can easily be removed with the ‘re-touch’ function. The ‘confirm roots’ function provides a further quality check, where unwanted objects can be deleted with a single mouse click. EZ-Rhizo is particularly good at dealing with low contrast because it allows the user to adjust the threshold for ‘black and white’ conversion. Choosing an appropriate threshold also removes the ‘haloes’ occasionally created by a high density of root hairs. The exact determination of the root borders is not crucial for the determination of RSA parameters, as EZ-Rhizo operates on the principle of skeletonization. An obvious limitation of this algorithm is that EZ-Rhizo does not measure the width of the main and lateral roots.

Plant size and complexity.  The size of the plants is immaterial for RSA analysis with EZ-Rhizo, as long as the uploaded picture displays the entire root system with sufficient resolution. On the 120 × 120-mm Petri dishes used in this study roots reached the bottom of the plates between 10 and 18 days after germination (depending on the accession), thus limiting the analysis to young plants. Larger plates can be used for the analysis of older plants or for faster growing species. An example of root detection with EZ-Rhizo from a 3-week-old A. thaliana plant grown on a 240 × 240-mm Petri dish is shown in Figure S3. The complexity of the root system is the only factor limiting RSA analysis in EZ-Rhizo, with respect to both time and accuracy. EZ-Rhizo measures uneven root paths and an unlimited number of higher order lateral roots with high speed and accuracy, but the program may make wrong decisions on path continuation when it encounters crossroads during root path tracking. In most cases the shorter end of the cross will be detected as an additional lateral root. Although the problem can be reduced by providing sufficient space for individual roots in the growth system, intersections of lateral roots with the main root or other lateral roots cannot be completely avoided in older plants grown on a two-dimensional surface. In this first version of EZ-Rhizo, the root segments concerned have to be detached and re-connected, or re-drawn using the ‘re-touch’ function. The cursor-aided brush application and subsequent re-skeletonization facilitate the operation and the final display, and confirmation of each root enables visual quality control. Nevertheless, editing of individual lateral roots requires some time if the RSA is very complex, and will invariably introduce some subjectivity into the analysis. We provide examples of root systems differing in complexity and occurrence of root intersections in the Supplementary material, and discuss the time requirements and effects of extended user input on individual RSA parameters (Figure S4, Tables S2 and S3 and Appendix S1).

Future developments

Future software improvements envisaged by our group include a number of different aspects from image detection to data analysis. Firstly, we aim to achieve automatic root path tracking across root intersections. In general, macroscopic image analysis cannot make decisions on individual root paths if it is impossible to make such decisions without microscopic tissue analysis. Nevertheless, our bare eye is often able to follow individual root paths by integrating the spatial and temporal ‘history’ of the root features. In this first version of EZ-Rhizo, the user has to apply such ‘knowledge’ by means of a simple manual editing function (see above). We are currently working on machine-learning algorithms to equip the program with memory-based knowledge. The automatically established database allows easy access to ‘memory’ in the form of data on the same root system from previous time points, and hence the EZ-Rhizo platform is well suited to tackle this challenge. Additional analysis modules are being developed to read RSA parameters from text files to reconstitute the root architecture for growth modelling purposes, and to predict root architecture under varying conditions in a database-supported machine learning process. Another obvious development is to create a publicly available database for RSA, where users can deposit their own databases and mine the entire data set. EZ-Rhizo provides a convenient prototype of such a depository with respect to both parameter definition and database structure. Information and downloads of future versions will be disseminated through the EZ-Rhizo webpage (http://www.ez-rhizo.psrg.org.uk), which includes a web-based forum group for user assistance and feedback.

Applications

EZ-Rhizo is suitable for investigating a wide range of biological questions. It will be of particular use in the areas of functional genomics, breeding and QTL, and predictive plant performance. EZ-Rhizo facilitates (i) root detection, (ii) quantification of output, and (iii) data storage and mining, and is therefore well suited for the phenotypic description of individual plant species, accessions and mutants grown under varying nutritional and environmental conditions. Many RSA phenotypes are already eminent at early stages of development, reducing the time required for manual editing to a minimum. Large screens for RSA mutants call for a fully automated analysis, which will only be available in future versions (see above). However, the current version of EZ-Rhizo can already be used without manual editing for mutant screens, as the detection of plants that significantly differ from the average population does not require identification of the crucial root component. Phenotypic characterization of RSA in candidate plants should subsequently be carried out with extended user input.

Natural variation of RSA in A. thaliana

Analysis of 23 A. thaliana accessions with EZ-Rhizo has further increased our knowledge about the genotypic plasticity of RSA. The study underscored the importance of measuring different RSA parameters to fully describe natural variation, and identified novel determinants of natural RSA variation (e.g. main root angle and lateral root length). The PCA plots we obtained can now be used to identify parental lines for QTL analysis based on distinct RSA parameters. EZ-Rhizo also unravelled spatial lateral root patterns, which could be correlated with the specific environment of the accessions. For example, Pyl1 presented few, but long, lateral roots only in the basal part of the main root. This extreme RSA is likely to be an adaptation to the sandy environment of this ecotype (Le Pyla, France), where long lateral roots in the upper part of the root anchor the plant in the sand, and the long main root reaches water and nutrients deep in the soil. Furthermore, using EZ-Rhizo we were able to evaluate how growth and growth rates of different root types change with development. Our finding that the main root growth rate increases between days 3 and 9 differs from a previous assessment of main root growth in A. thaliana accessions (Beemster et al., 2002). We suspect that the difference in this case relates to the lower supply of nutrients and light in our study, which delayed development so that the main phase of root growth acceleration could be captured over the assessed period of time. In accordance with this interpretation, we found that even in our conditions very fast-growing accessions, such as WS, exhibited less acceleration of main root growth between days 3 and 9 than during previous days. Interestingly, lateral root length of the same plants still increased exponentially over the entire period assessed here. These findings call for much more detailed kinetic analyses of root growth in the future. EZ-Rhizo presents a fast technical solution to carry out such analyses, and thus provides a basis for the identification of genes controlling speed and prioritization of root growth during plant development.

Experimental procedures

Software development

The EZ-Rhizo application was developed on, and for, the Microsoft® Windows® operating system, in the C++ language using Microsoft® Visual Studio® 2005. The software–database interaction was implemented using the EasyPHP (http://www.easyphp.org) and MySQL (http://www.mysql.com) application programming interfaces.

Downloading EZ-Rhizo

EZ-Rhizo software (for Windows operating systems) is freely available to the scientific community. It can be easily downloaded from the EZ-Rhizo website (http://www.ez-rhizo.psrg.org.uk) and installed using our installation wizard. The website also contains a tutorial, a video demonstration and a forum for user feedback.

Plant material and growth conditions

Arabidopsis thaliana seeds were surface sterilized (2.5% sodium hypochlorite, 0.1% Tween-20) for 5 min, rinsed five times with sterile water and placed in darkness at 4°C for 3–4 days to synchronize germination. Four seeds were then sown in 120 × 120-mm square Petri dishes containing 30 mL of half-strength MS medium (Murashige and Skoog, 1962), adjusted to pH 5.6, and supplemented with 1% sucrose and 1% agar Type A (A4550; Sigma-Aldrich, http://www.sigmaaldrich.com). Petri dishes were sealed with parafilm and placed vertically under the light source (16 h day−1 at about 100 μE) at a constant temperature of 22°C. Synchronous germination across accessions was verified visually and occurred 48 h after seed sowing. For image acquisition, plates were placed in a scanner (HP scanjet 4500c; http://www.hp.com) and immediately afterwards returned to the growth chamber.

Statistical analysis

The RSA parameters were retrieved from the database using the EZ-Rhizo query menu, and were saved in the csv file format. Graphs and statistical analysis were performed in Excel using the xlstat program (Addinsoft, 2008).

Acknowledgements

We are grateful to Dr Olivier Balédent (Department of Imaging and Biophysics, University Hospital, Amiens, France) for fruitful discussions and advice, to the Socrates and Leonardo programs for enabling undergraduate students Michael Cornière and Patrick Lageron from ESIEE (Ecole Supérieure d’Ingénieur en Electronique et Electrotechnique, Amiens, France) to take part in the software development, and to Prof. Richard Cogdell (FBLS) for his help with securing funding for KZ from the Pfitzer Bower Fire Fund. PA and RP were supported by the BBSRC. Many thanks also go to the members of the Blatt/Amtmann laboratory who tested the software and provided valuable feedback.

Ancillary