The Plant Journal (2019) 98, 1145–1156 doi: 10.1111/tpj.14297 TECHNICAL ADVANCE MyROOT: a method and software for the semiautomatic measurement of primary root length in Arabidopsis seedlings Isabel Betego n-Putze1,‡, Alejandro Gonzalez2,‡, Xavier Sevillano2, David Blasco-Escamez1 and Ana I. Can~ o-Delgado1,*,† 1 Department of Molecular Genetics, Center for Research in Agricultural Genomics (CRAG) CSIC-IRTA-UAB-UB, Campus UAB, Bellaterra (Cerdanyola del Valle s), 08193, Barcelona, Spain, and 2 GTM- Grup de Recerca en Tecnologies Me dia, La Salle, Universitat Ramon Llull, 08022, Barcelona, Spain Received 20 September 2018; revised 18 February 2019; accepted 19 February 2019; published online 27 February 2019. *For correspondence (e-mail ana.cano@cragenomica.es). ‡ These authors contributed equally to this work. † Present address: Department of Molecular Genetics, Centre de Recerca en Agrigeno mica (CRAG) CSIC-IRTA-UAB-UB, Campus UAB, Bellaterra (Cerdanyola del Valles), E-08193, Barcelona, Spain. SUMMARY Root analysis is essential for both academic and agricultural research. Despite the great advances in root phenotyping and imaging, calculating root length is still performed manually and involves considerable amounts of labor and time. To overcome these limitations, we developed MyROOT, a software for the semi- automatic quantification of root growth of seedlings growing directly on agar plates. Our method automati- cally determines the scale from the image of the plate, and subsequently measures the root length of the individual plants. To this aim, MyROOT combines a bottom-up root tracking approach with a hypocotyl detection algorithm. At the same time as providing accurate root measurements, MyROOT also significantly minimizes the user intervention required during the process. Using Arabidopsis, we tested MyROOT with seedlings from different growth stages and experimental conditions. When comparing the data obtained from this software with that of manual root measurements, we found a high correlation between both methods (R2 = 0.997). When compared with previous developed software with similar features (BRAT and EZ-Rhizo), MyROOT offered an improved accuracy for root length measurements. Therefore, MyROOT will be of great use to the plant science community by permitting high-throughput root length measurements while saving both labor and time. Keywords: root length, software, root phenotyping, high-throughput image analysis, Arabidopsis thaliana, technical advance. INTRODUCTION The root, responsible for anchoring the plant to the soil, is Arabidopsis thaliana (Arabidopsis), the primary root has a an essential organ for overall plant growth and develop- simplified anatomy that makes it very amenable for genetic ment. Characterization of different root traits is therefore and microscopic analysis (Dolan et al., 1993; Ishikawa and important not only for understanding organ growth, but also Evans, 1995; Iyer-Pascuzzi et al., 2009). Different root cell for evaluating the impact of roots in agriculture (Kuijken lineages are derived from a group of stem cells located at et al., 2015). As such, generating tools for precise, high- the root apex. Here, the stem cell niche is formed by a few throughput phenotyping and imaging of roots is essential (three to seven) quiescent center cells that occasionally for plant research and agriculture. Phenotyping facilities, divide asymmetrically to renew themselves and to form such as the ones available in the European Plant Phenotypic daughter stem cells. From the root apex, these cells Network (http://www.plant-phenotyping-network.eu/), have actively divide in the meristematic zone and, before exiting started to implement tools for mass screening of roots. the cell cycle in the transition zone, continue to elongate Roots provide necessary structural and functional sup- and differentiate in spatially separated regions of the root. port for incorporation of nutrients and water from soil. In In this way, primary root growth is determined by the © 2019 The Authors. 1145 The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.  n-Putze et al. 1146 Betego balance between cell division and cell elongation within Despite the numerous tools for root analysis, these lack the different zones of the root (van den Berg et al., 1997; the capability to precisely measure the primary root length Beemster and Baskin, 1998; Verbelen et al., 2006; Takat- of seedlings. This approach is key for genetic, physiologi- suka and Umeda, 2014). cal and developmental studies in the plant model Ara- The most straightforward symptom of abnormal root growth bidopsis and is often done manually. Here, we present or development can be identified by examining the length of MyROOT, a software capable of semiautomatically calcu- the primary root in seedlings. Abnormalities in length can usu- lating primary root length in a fast and user-friendly man- ally be observed and measured just 5–6 days after germination ner, and that is able to adapt to different imaging and (DAG), which still reflect their embryonic origin (Ju € rgens et al., experimental conditions. By automatically identifying the 1995). Growth defects in the primary root of seedlings are not scale and precisely detecting all individual roots and hypo- only consistent with overall growth defects, but also persistent cotyls growing on an agar plate from a JPEG image, this along the entire plant life cycle (Benfey et al., 1993; Potuschak software simplifies and minimizes user intervention during et al., 2003; Gonzalez-Garcia et al., 2011). Indeed, Arabidopsis the calculation of root length. MyROOT merely requires root analyses were the foundations for multiple genetic screens the user to define the region in which the seedlings are that ultimately led to the identification of several key regulators placed on the plate, and then subsequently operates in a of plant growth and development (Benfey et al., 1993; Hauser semiautomatic fashion. We show that MyROOT can be et al., 1995; Can~ o-Delgado et al., 2000; Mouchel et al., 2004; used both in low-scale and high-throughput experiments Ubeda-Tomas et al., 2008). due to the incorporation of a batch-processing option for Root analysis of young seedlings offers direct information automatic processing of several images without losing its regarding overall plant growth and viability. Despite impor- accuracy. tant advances in plant imaging techniques such as micro- As a proof of concept, MyROOT software was used for scopic visualization (Pfister et al., 2014; Gonzalez-Garcia root length measurement of Arabidopsis wild-type and et al., 2015; Lobet, 2017), the root length of seedlings grow- brassinosteroid (BR)-signaling mutants grown under con- ing in agar plates is generally measured by manually indicat- trol conditions, exogenous BR hormones treatment and for ing the position of each seedling or manually tracking each plants grown under osmotic stress conditions (Fabregas root using the ImageJ software (https://imagej.nih.gov/ij/). et al., 2018). MyROOT software is available at https:// For this reason, the development and use of methods that www.cragenomica.es/research-groups/brassinosteroid- enable automatic and accurate analysis of a large number of signaling-in-plant-development. roots represents a step forward for high-throughput root analysis. Automatic analysis of root system architecture is RESULTS just beginning to be implemented, and novel methods based MyROOT is a software for high-throughput analysis of on acquiring, processing, and obtaining quantitative data root length from root images are now available. The available root softwares are designed for different Most root studies begin with an overall determination of purposes. Some of these are applied to crop phenotyping root growth as estimated by manual, laborious and time- and are mainly focused on traits related to root architecture consuming measurements. To address this limitation, we such as branching or biomass (Le Bot et al., 2010; Lobet developed a semiautomatic and non-invasive software for et al., 2011; Nagel et al., 2012; Pound et al., 2013; Pace the high-throughput measurement of root length. This et al., 2014; Kuijken et al., 2015; Wu et al., 2018). Con- method is implemented in MATLAB as an automatic tool versely, other root softwares are oriented for the analysis of named MyROOT (Figure 1a). It is based on pictures of plant model species as Arabidopsis (Armengaud et al., whole agar plates on which young seedlings are growing 2009; French et al., 2009; Yazdanbakhsh and Fisahn, 2012; vertically on the surface, and implements novel algorithms Slovak et al., 2014). Most of these can reliably measure dif- capable of separately detecting the root and the hypocotyl ferent root traits (primary root length, lateral roots, etc.) of each individual seedling and estimating a hypocotyl (Arsenault et al., 1995; Le Bot et al., 2010; Clark et al., 2013; curve based on the detection of some hypocotyls Ristova et al., 2013; Slovak et al., 2014; Cai et al., 2015). The (Figure 1b–g). analysis usually requires intensive user intervention to set MyROOT detects and measures root length by following the optimal parameters for root detection and to identify a series of steps (Figure 1b–g, Box 1–3 and Video S1). First, the individual roots (Armengaud et al., 2009; French et al., a digital image of the plate containing the growing seed- 2009; Clark et al., 2013). Despite time-consuming manual lings is taken and used for the analysis (Figures 1b and tracking of each single root, ImageJ is often used as a tool S1). The image has to include a ruler (at least 1 cm long) for primary root length quantification. A summary of the placed on top of the plate. From the JPEG image, the soft- existing softwares for root analysis and its main features is ware: (i) detects 1 cm of the ruler to automatically compute given in Table S1. the scale and calculate the equivalence between pixels and © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156 MyROOT software for plant root length measurement 1147 millimeters (mm; Figure 1c); (ii) generates a binary mask tape that is placed on the surface of the Petri dish. This from the manually selected area that allows for root seg- allows the measurements to be completely independent mentation (this separates those pixels that belong to a root from the specific characteristics of the image capture sys- from those of the background) (Figure 1d); (iii) measures tem. The first step for detecting the ruler is based on its the length of the roots through a root-tracking process color contrast with the background. By computing the ver- (Figure 1e); (iv) computes a regression curve based on the tical and horizontal profiles of the image, the algorithm is detection of the hypocotyls to identify the starting point of designed to explore the entire image in search of a white each root (Figure 1f); (v) measures the root length again patch (Figure S2a). As the border of the plate has a similar from the root tip to the end of the hypocotyl (Figure 1g); color contrast with the background, a median filter is and (vi) exports the measurements and the generated applied to reduce the border effect. The maximum values masks to a new folder. Finally, the results are saved in: (i) a in the filtered profiles define the image area where the Microsoft Excel spreadsheet or a .txt file in which each root white patch is present. Next, the resulting area is further is identified by an ID tag, length value and a descriptive cropped (Figure S2b) and processed (Figure S2c–e). By text label introduced by the user; (ii) an image showing the applying a threshold based on Otsu’s algorithm (Otsu, detected and measured roots; (iii) MATLAB variables 1979), the black lines representing cm and mm marks are including the intermediate data such as hypocotyl position not filtered out (Figure S2b). Finally, a horizontal profile of and the detection curve that was generated while quantify- this binary image is generated (Figure S2d) in which the ing root length and (iv) a RSML file so the images can be pixel-mm equivalence is defined as the difference between analyzed with other compatible softwares (Lobet et al., consecutive local maxima (Figure S2e). In case that there 2015). is not ruler tape in the image to analyze, MyROOT includes One of the advantages of MyROOT is that it allows the the option of manually indicating the correspondence user to supervise the different steps of the process as the between pixels and millimeters. results of each step are displayed before executing the fol- The core of the whole method is the root extraction lowing one. This feature enables the user to modify the and measurement process. To extract roots, the user different parameters (e.g., segmentation thresholds for must first manually define the area in which roots are ruler and root detection, and model for hypocotyl detec- present (note: only one row of seedlings should be tion, etc.) at any point in the process to take into account included when defining the area). Then, with just a few different image conditions. Nonetheless, default parame- mouse clicks from the user, a binary mask is generated ter values have been set for satisfactory operation on a that allows root segmentation. This later leads to the wide range of images for pre-defined acquisition condi- identification of individual roots through a root tracking tions (see Experimental procedures). Furthermore, the process, and finally allows the individually identified position of any hypocotyl that is not automatically roots to be measured (Figure 2). The root segmentation detected can be manually indicated, and undesired roots process can be divided into four main steps: (i) color nor- can be manually removed from the results before saving. malization (Figure 2a), (ii) ridge detection (Figure 2b), (iii) In addition, MyROOT incorporates a batch processing root tracking (Figure 2c), and (iv) root identification (Fig- option for an automatic high-throughput analysis in which ure 2d). During the color normalization step, the image is the different parameters are set for the first image from a processed and a global working framework is set (i.e., all specific folder and automatically applied in the rest of images going through this process become color- these. balanced and have the same lower and higher white val- In summary, by determining the pixel-millimeter equiva- ues; Figure 2a). This allows the user to manage different lence and detecting seedling morphology (roots and hypo- initial conditions (illumination, color, and saturation, etc.) cotyls) from an image of a seedling-containing agar plate, while continuing with the same subsequent steps of the MyROOT offers a valuable analytical tool for precisely pipeline. In the next step, a ridge (i.e., white contrasted measuring root growth in a semi-automatic manner. As area) detector identifies roots based on their contrast such, this software clearly provides a solution to the timely with the background (for this, the level of whiteness is task of manually quantifying root length. irrelevant; Figure 2b). After the detection step, a final mask is generated for tracking the roots. Due to the lin- Root detection and measurement process ear disposition of the roots in the plate, we employed a MyROOT has been developed for the high-throughput, bottom-up tracking approach. As such, tracking starts at accurate, and non-invasive measurement of root length the end point of each root and continues upward, row by from seedlings growing in agar plates. In this respect, the row, until the hypocotyl detection curve is found (Fig- three most crucial steps are to precisely determine the ure 2c). Finally, the tracking of each root makes it possi- scale, identify the roots, and measure their length. The ble to identify which pixels correspond to which root scale information is obtained from a piece of measuring (Figure 2d). © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156  n-Putze et al. 1148 Betego (a) (b) (d) (e) (f) (c) (g) Figure 1. The graphical interface and steps of MyROOT. (a) The graphical user interface of MyROOT is organized into seven sections: 1. Input image information, 2. Root extraction parameters, 3. Hypocotyl detection parameters, 4. Manual removal of roots, 5. Visualization of the image and the different detection steps, 6. Saving parameters, and 7. Batch processing (b) The input image required for analysis is a picture of the square plate in which the aligned seedlings are growing. By using information from this image, MyROOT performs the following steps: (c) Identification of the ruler to determine the scale (i.e., the equivalence between pixels and millimeters). (d) Root segmentation to identify the seedlings. (e) Root tracking to measure the roots. (f) Hypocotyl detection to identify the hypocotyls and separate them from the roots. (g) Root measurement to quantify the length of individual seedlings (i.e., the distance from the root tip to the end of the hypocotyl). © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156 MyROOT software for plant root length measurement 1149 Once the root tracking process has been completed, with manual measurements, thereby supporting the use of each individual root is measured based on previous posi- MyROOT for root length analysis in seedlings in different tions saved in the historical record. Specifically, root length growth stages and experimental conditions. is calculated in pixels by adding the distances between We also evaluated the time required by MyROOT to previous consecutive points and then applying the previ- determine root lengths, and compared it with the time ously calculated pixel-mm equivalence. Next, a refinement needed for manual measurements using ImageJ, as it is process is applied in which the shortest roots, which are widely used for the analysis of a low number of plates as a often associated with noise, are discarded. By default, routine task in many plant biology laboratories. Impor- MyROOT discards any root measurement shorter than 30% tantly, we found that MyROOT reduces the time required of the longest one. However, this percentage can be manu- to measure one plate by approximately half (Figure S3). ally chosen by the user if need be. A second filter is then When using MyROOT for high-throughput experiments, applied to keep those roots that terminate close to the pre- the analysis of a batch of images can be done automati- viously calculated hypocotyl curve. If a root surpasses the cally after setting the optimal parameters for adapting to hypocotyl curve, it is cut at this level. Finally, unique the imaging conditions of the experiment. The time numeric identifiers (ID) are assigned to all roots that are required for the analysis of one row of seedlings of differ- not filtered out during processing. ent plates was evaluated. MyROOT spends approximately As two roots can be located so close to one another that 1 min per image when the hypocotyl detection is not per- they cannot be detected as individual roots, MyROOT was formed and 1 min more if it is performed. Importantly, this trained with the following characteristics: (i) when a split process is completely automatic and does not require the occurs and a current root matches more than one detec- user intervention. The accuracy of the batch processing tion (blue circle in Figure 2c), a new root sharing the same was evaluated by comparing the results of the analysis of historical record is created, and (ii) when a fusion occurs 10 different plates using MyROOT by single upload fol- and two roots match a single detection (yellow circle in lowed by individual setting of the optimal parameters and Figure 2c), the shortest root is eliminated from the root set using MyROOT for a batch analysis of all of them automat- and added as a sub-root of the longest one; therefore indi- ically (Figure S4). The correlation obtained between both cating the root length of the primary root which is longer methods was positive (R2 = 0.996, Pearson’s r = 0.9981), than the lateral roots. therefore confirming that the batch processing option can To validate our software, we compared root length mea- be performed without losing accuracy in the final measure- surements obtained using MyROOT with manual measure- ments. ments performed using ImageJ. We compared the root Hypocotyl detection length values of different experiments. First, 6-days-old seedlings of wild type and BR-related mutants grown in One of the main advantages of MyROOT is its ability to control and in osmotic stress conditions (data published in identify hypocotyls of growing seedlings. This characteris- Fabregas et al. (2018), n > 600, Figure 3), and second, the tic is important for accurately determining the start point same seedlings over 6 consecutive days (from three to of each root. The hypocotyl detection process is based on eight DAG; n > 116). We obtained a positive correlation visual features extracted (appearance and color) from the between the measurements with both methods image. These features were used to generate a hypocotyl (R2 = 0.997, Pearson’s r = 0.9985). These results indicated model by introducing 1259 hypocotyls of seedlings of dif- that measurements made using our software coincided ferent ages and characteristics and 7915 samples with Figure 2. Root extraction method. (a) Colors are normalized in the area where roots (a) (b) are present, and white roots are detected. (b) Seg- mentation is performed by applying a ridge detec- tor. (c) Starting at the root tip, the roots are tracked using a bottom-up approach. (d) Each root is mea- sured using its historical recorded tracking, and root length is calculated by taking into account the pixel-millimeter equivalence. (c) (d) © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156  n-Putze et al. 1150 Betego three different models that differ in the type of feature they 30 use for describing hypocotyls: only color information, only Root length by MyROOT (mm) appearance information (via HOG features), or both types of information (HOG + color). 20 Upon analyzing the precision-recall curve of each model, we found the HOG + color model to be the most robust (Figure 4b). In the case of FPPI, the lowest miss rate was also found when using the HOG + color model (Figure 4c). 10 These results indicated that when considering both color R2 = 0.997 and appearance (i.e., the HOG + color model), a higher Pearson’s r = 0.9985 number of hypocotyls were identified than when using 0 only one of the features. Therefore, this validates our 0 10 20 30 MyROOT method because it incorporates both HOG and Root length by ImageJ (mm) color information. Next, we evaluated the influence of different regression Figure 3. Validation of root length measurements. curve models on the root measurement refinement used to Correlation of root length measurements using MyROOT (y axis) and Ima- geJ (x axis). Each point corresponds to a different experiment (n > 20 in set up the limits of individual roots (Figure 4d). To create each one): time course data from 3 DAG to 8 DAG seedlings (grey) and BR- these curves, a regression upon the detected hypocotyls related mutants in control and osmotic stress conditions (black, Fabregas was performed. To define which regression model gives et al., 2018). Errors bars indicate the standard error. For the time course experiment, seedlings that were not measured by MyROOT in at least four the better fit, we tested different polynomial models that time points were discarded. were evaluated in terms of the average distance (in pixels) between the real hypocotyl position and the point of inter- section between the root and the regression curve (Fig- ure 4d). The results indicated that when using a hypocotyl background information (see Experimental procedures). regression curve of order 4, a good balance between accu- The learned model is able to determine whether a given racy and flexibility that is able to account for small changes sample is a hypocotyl or not. To extract visual features, we in hypocotyl position is reached. Therefore, we chose to implemented the histogram of oriented gradient (HOG) employ this regression curve in our software. (Dalal and Triggs, 2005) method. The HOG method is The use of the hypocotyl detection method permits the based on the orientation of the contours in the image, and fine identification of the starting point of the root (Fig- generates a histogram that represents the appearance/ ure 5a–d). Whereas, depending on the user judgement, shape of the sample. For extracting color features, color this option can be skipped, therefore reducing the time of distribution histograms representing the amount of color the measurement process but losing accuracy in the final in a given sample area are used (Figure 4a). To train the root length results (Figure 5e). We compared the root model, we implemented a linear support vector machine length measurements of two plates, one containing seed- classifier that uses appearance and color features from the lings with standard hypocotyls (Col-0 wild type plants) and hypocotyl images. This classifier generates the best hyper- other with shorter hypocotyls and roots (Col-0 wild type in plane that classifies samples as positive (hypocotyls) and osmotic stress conditions), using and skipping the hypoco- negative (no hypocotyls) examples. During the hypocotyl tyl detection step (Figure 5a–d). In both plates, when using detection stage, the sliding window approach (Glumov the method, only the primary root is measured (Figure 5a, et al., 1995) is used to perform an exhaustive search for c), whereas when it is not used, some parts of the hypoco- hypocotyls. Finally, by keeping the highest scored win- tyl are measured too (Figure 5b,d). When not using the dows as true positives, polynomial regression is used to hypocotyl detection method, we only found statistically define a curve that passes through all the detected hypoco- significant differences in the root length of the shorter tyls. Although the user can manually insert the location of seedlings, as the proportion of hypocotyl length measured the hypocotyls, this curve enables the position of unde- significantly increased the overall root length measure- tected hypocotyls to be estimated, and therefore corrects ment (Figure 5e, n > 30). These results highlighted the the curve tracing. The intersection between the hypocotyl importance of the hypocotyl detection process for accu- detection curve and each root is used to define the root rately measuring the root length, but also point to just start point. using it when the experiment requires high precision. We first evaluated our hypocotyl detection process in Comparison of MyROOT with available root softwares terms of different hypocotyl detection models. Both the precision-recall curve (Figure 4b) and the number of false The choice of software is usually based in a balance positives per image (FPPI; Figure 4c) were calculated for between the appropriateness of the characteristics of the © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156 MyROOT software for plant root length measurement 1151 (a) Candidate Image Block window gradient descriptor HOG/Color candidate descriptor (b) (c) 1 1 0.8 0.8 Precision 0.6 Misrate 0.4 HOG+Color – 0.6 0.6 HOG+Color – 66.6% 0.2 HOG – 0.6 HOG – 67.6% Color – 0.5 Color – 73.1% 0 0.4 0 0.5 1 0 5 10 Recall FPPI (d) 50 Distance to hypocotyl 40 1. a + bx curve (pixels) 2. a + bx + cx2 30 3. a + bx + cx2 + dx3 20 4. a + bx + cx2 + dx3 + ex4 5. a + bx + cx2 + dx3 + ex4 + fx5 10 6. a + bx + cx2 + dx3 + ex4 + fx5 + gx6 0 7. a + bx + cx2 + dx3 + e·sin(x) 1 2 3 4 5 6 7 Model Figure 4. Hypocotyl detection method and validation. (a) Scheme of the hypocotyl detection method. A candidate window is defined as a square area inside the image. To describe a candidate, appearance/shape (HOG) and color information are extracted. Appearance information is extracted to calculate the gradient of the image (i.e., the direction of the contours within the image at each pixel). Histograms of Oriented Gradient (HOG) and the histograms of color are calculated over regular spaced, non-overlapping cells inside the candidate window (forming the block descriptor). Finally, all color/HOG cell histograms are concatenated to obtain the candidate window description. (b) Precision-Recall curve for three different models of hypocotyl detection (HOG, Color and HOG + Color). The curve is obtained by changing the threshold that defines the frontier between positive and negative samples. For each threshold, the precision (well classified ratio) and the recall (poor classified ratio) were cal- culated. The area under the curve represents the robustness of the classifier, with a higher value indicating greater robustness (a higher well classified ratio to poor classified ratio over the entire range of the classifier). (c) False Positives Per Image (FPPI) curve for three different models of hypocotyl detection (HOG, Color and HOG+Color). The curve plots the miss rate against the FPPI. In this way, the average miss rate over a specific FPPI range (1–10) represents the sensitiv- ity of the classifier to not miss good samples and keep the false positive ratio low. (d) The average distance in pixels between the real hypocotyl position and the point of intersection between the root and the polynomial regression curves, for polynomial regression curves of orders 1–6 and an extra model including a sine component. Error bars indicate the standard error. software to the experimental design, the accuracy of the comparison with BRAT (15.08 mm and 1.77 mm) and to measurements obtained and the time required for the anal- EZ-Rhizo (2.67 mm and 2.42 mm) (Table S2). ysis (Table S1). We compared the accuracy and the time Root detection on the plate was similar between spent by MyROOT with BRAT and EZ-Rhizo, the two most MyROOT (> 90%) and EZ-Rhizo (> 96%), yet MyROOT pro- similar software tools, in the quantification of primary root vided more accurate measurements (Figure S5 and length of two independent plates containing Arabidopsis Table S2). In addition, due to the incorporation of the seedlings (Figure S4 and Table S2). hypocotyl detection method, MyROOT requires less user We first compared the root length obtained with the intervention to clearly indicate the roots, this is reflected in three softwares with the ImageJ results. The absolute dif- a reduction in the time spent for the analysis. We spent ference of mean root length between the measurements around 3 min to analyze each plate when using MyROOT, obtained with the softwares and with ImageJ show that 0.5 min with BRAT and 15 min with EZ-Rhizo (Table S2). MyROOT differs 1.39 mm and 0.22 mm for plates 1 and 2 Some seedlings on the plates analyzed had overlapping respectively. With regard to this parameter, our results hypocotyls. We found that MyROOT was able to identify indicated the better performance of MyROOT in these and correctly indicate the shoot root junction in © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156  n-Putze et al. 1152 Betego (e) (a) (b) (c) (d) Figure 5. Evaluation of hypocotyl detection method for the root length measurements using MyROOT. (a–d) Qualitative analysis of the hypocotyl detection method in two different images. (e) Root length measurement of seedling grown in two different plates (showed in (a–d)) using and not using the hypocotyl detection method. Different letters indicate statistically significant differences (P-value <0.05; Student’s t-test, n > 30). cases with higher precision than EZ-Rhizo and BRAT, (Figure S5 and Table S2). This can be explained because which presented a more basic algorithm for the hypocotyl of the limited control that BRAT offers for the adjustment detection based only on color and shoot border curvature of its internal parameters by the users to adapt to different information, respectively. These observations highlight the imaging conditions, which may result in low root detection utility and importance of the hypocotyl detection method rates (Table S2). Conversely, MyROOT easily allows the incorporated in MyROOT and its capacity to identify the modification of different thresholds (scale and root mask individual seedlings and the precise starting point of each threshold and hypocotyl detection method) to define the root. optimal parameters for the analysis. Overall, MyROOT fills a specific gap in root phenotyping Another novel feature of MyROOT compared with exist- by allowing a precise, fast and semi-automatic quantifica- ing softwares is the ability to automatically identify the tion of primary root length of seedlings on a plate and a scale by detecting a ruler over the imaged plate; therefore batch of plates. being able to automatically adapt to different imaging con- ditions and settings; and therefore being able to work inde- DISCUSSION pendently from specific hardware set ups (Table S1). In this study, we presented MyROOT software for the Importantly, MyROOT is the only software that uses quantification of plant root length. Despite the existence of machine learning-based algorithms to identify Arabidopsis other software (Table S1), MyROOT covers a gap in analy- hypocotyls, which allows its identification in different sis of Arabidopsis seedling not previously addressed. The mutants and experimental conditions even when they are determination of root length in such accuracy is key in the overlapped or present different morphologies. In excep- study of plant growth and developmental processes, in tional cases in which the hypocotyls are not automatically which small root differences can lead to the identification identified, or when the software is used for the analysis of of important genes (Benfey et al., 1993; Li et al., 2001; other plant species, they can be manually indicated by sim- Mouchel et al., 2004; Rodrigues et al., 2009). We found that ply pointing over them. the implementation of primary root phenotyping algo- In addition to the automatic processing of several rithms in a platform using independent, semi-automatic images, MyROOT allows the user supervision during all and user-friendly software to accurately measure root the process. It displays intermediate results during the length achieved by using MyROOT will replace the current analysis and allows the modification of different parame- manual and time-consuming tools such as ImageJ. ters to get the optimal results. Despite that MyROOT only MyROOT gives the most precise root length measure- indicates the primary root length, the output is also saved ments when compared with similar software BRAT and EZ- in rsml software that allows its compatibility with other Rhizo (Figure S5 and Table S2). Importantly, the accuracy software that can determine other aspects of root architec- is maintained both in the manual and in the batch process- ture such as curvature or branching. ing mode of MyROOT (Figures 3 and S4), being therefore Overall, this article advances the utility of MyROOT for suitable for high-throughput experiments. Under our determination of Arabidopsis root length in a precise, fast experimental conditions, only BRAT operates in a faster and simple manner. It incorporates powerful algorithms to manner, whereas the final results are less accurate identify scale and seedlings over standard images of agar © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156 MyROOT software for plant root length measurement 1153 plates. These are incorporated in a user-friendly graphical proving that this software can be used within the research interface that allows supervision and manipulation during community to perform high-throughput experiments in a the different steps of the analysis. less time-consuming manner. Finally, MyROOT is a modularly designed software con- sisting of a group of specialized algorithms able to detect EXPERIMENTAL PROCEDURES and analyze measuring tape, detect roots, track roots in a Plant material and growth conditions bottom-up fashion and detect hypocotyls. Therefore, any improvement to any of these components, or new algo- Arabidopsis thaliana Col-0 seeds were surface sterilized with a 5- rithms to determine other features, can be easily included min incubation in 1.5% sodium hypochlorite, followed by five in subsequent versions of MyROOT. Examples of future washes in distilled sterile water. Seeds were stratified for 48 h at 4°C in the dark to synchronize germination. Seeds were sown in improvements that could be included are the development 12 9 12-cm plates containing half-strength Murashige and Skoog of daily growth-monitoring algorithms that permit the (½MS) medium without sucrose and supplemented with vitamins. detection of abnormal root growth patterns, analysis of root Seeds were distributed individually in the plate in two rows with system architecture beyond the primary root, and identifica- around 15 seeds per row. The plates were incubated for 3–8 days tion of hypocotyls from other plant species. In the future, vertically oriented under long day conditions (16 h of light and 8 h of dark) at 22°C and 60% relative humidity. upgraded versions of our software could consist of a com- pletely automatic operation connected to high-throughput Statistical analysis facilities for large-scale characterization of root traits. The comparison between the root length measurements using CONCLUSION MyROOT and ImageJ was evaluated by performing a regression curve and calculating the Pearson correlation coefficient. The com- MyROOT is a software capable of semiautomatically mea- parison between the root length measurements using and not suring the length of the primary root of Arabidopsis seed- using the hypocotyl detection method was evaluated using Stu- dent’s t-test. This test was selected due to the unequal number of lings. It automatically recognizes the scale of the image, seedlings detected under each condition. and detects the hypocotyls and root tips from young seed- lings growing vertically in agar plates. This information is Plant Imaging and computer settings then used to accurately calculate the root length of each Images were taken with a D7000 Nikon camera. The pre-defined individual plant. This software was designed in such a way image acquisition conditions consist of placing the camera 50 cm that only a simple image of the plate is required for analy- above the plate with an illuminated support and the following set- sis. Importantly, MyROOT is even able to recognize hypo- tings: aperture 13, shutter speed 10, ISO 100 and Zoom 935 mag- cotyls of different ages and morphologies, and can nification. The plates were placed face down on a black surface and with a ruler (at least 1 cm long) horizontally positioned on top therefore be applied in a large range of experiments. (Figure S1). The images were saved in JPEG format (size between Here, our validation experiments demonstrated the high 2.5 and 2.7 MB per image). MyROOT and ImageJ were run in an precision of measurements made with MyROOT, thereby Intelâ CoreTM i7-6700 CPU computer. Box 1 Installation guide for MyROOT software 1 Execute MyAppInstaller_mcr.exe. 2 In the Root Analysis Installer window click the Next button. 3 In the Installation Options window select the folder where you wish to install the software. By default, the selected folder is C:\Program Files\La Salle – Universidad Ramon Llull\Root_Analysis. 4 Mark the option Add a shortcut to the desktop. 5 Click the Next button. 6 In the Required Software window select the installation folder for the MATLAB Runtime (by default C:\Program Files \MATLAB\MATLAB Runtime). 7 Click the Next button. 8 In the License Agreement window mark the option Yes. 9 Click the Next button. 10 In the Confirmation window click the Install button. 11 Copy the HypocotylDetection folder to the desktop (this folder is in the same folder as the .exe file used for the installation). 12 Open the software by clicking in the MyROOT desktop icon. © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156  n-Putze et al. 1154 Betego Box 2 Brief user guide for individual plates analysis using MyROOT software 1 Open the software in a PC. 2 Select and load the image to process by pressing the LOAD IMG button. Enter an image resize factor between 0 and 1 in the Scale edit box to reduce the size of the image and speed up the processing of high-resolution images. 3 Obtain the pixels-to-millimeters scale factor by clicking the Ruler ID button. If needed, edit the Ruler Threshold value to modify the sensitivity of the ruler detector and repeat step 3. If there is no ruler in the image, insert the correspon- dence between pixels and millimeters manually in the 10 mm equivalence box. 4 Start the root segmentation process by clicking the Root Mask button. Select the area where the roots are present by clicking in the image. Double click in one of the vertex to start generating the mask. In case the result is not satisfac- tory (e.g. oversegmented roots), modify the sensitivity factor in the Root Threshold edit box and repeat step 4. 5 Enter a value in the Root Length Threshold box to indicate the minimum percentage with respect to the longest root to be measured. 6 Start the root tracking and measurement process by clicking the Root Detection button. 7 Enter the path of the files containing the pre-trained hypocotyl detection models in the Hypocotyl Model Path edit box. 8 Optionally, to perform hypocotyl detection based on color descriptors only, check the Only Color checkbox, and to conduct a channel-wise color normalization process check the Norm Color checkbox. 9 Optionally, modify the threshold of the linSVM classifier by modifying the value in the edit box located next to the Hypocotyl Detection button. 10 Start the hypocotyl detection process by clicking the Hypocotyl Detection button. 11 If some of the hypocotyls were undetected, insert them manually by using the Add button. Click the Enter key and click the Root Refinement button to update root length measurements. 12 Remove the undesired roots from the measurement by typing the root identifier in the ID edit box and clicking the Remove button. Click the Visualize button to refresh the image presented on MyROOT’s visualization canvas. 13 Enter the path where you would like the results to be stored in the Results Path edit box. 14 Choose the type of data you want to save by checking the corresponding checkboxes. 15 Optionally, type an identification suffix that will be appended to the stored file names via the Root Label edit box. 16 Save the results by clicking the SAVE button. Box 3 Brief user guide for batch processing using MyROOT software 1 Load one of the image of the folder to process and select the optimal parameters for the analysis (steps 2 to 9 in BOX 2). 2 Enter the path where you would like the results to be stored in the Results Path edit box. 3 Choose the type of data you want to save by checking the corresponding checkboxes. 4 Optionally, type an identification suffix that will be appended to the stored file names via the Root Label edit box. 5 Indicate the name of the folder with the images to process (Name\) in the Folder Path box. 6 Click the Process button to start the analysis. 7 Select if using the hypocotyl detection method by clicking in the Yes or No button in the box that appears. Hypocotyl detection model Availability of data and materials The software was trained to identify hypocotyls by using 1259 MyROOT has been developed in MATLAB (version 8.3.0.532. positive examples (hypocotyls) and 7915 background and negative Natick, Massachusetts: The MathWorks Inc., 2014). It will be examples (parts of the image that did not contain hypocotyls). made available to the plant sciences community through the The positive samples corresponded to Col-0 wild-type, bri1-116, Plant Image Analysis website (plant-image-analysis.org; Lobet and a transgenic line overexpressing BRI1-GFP, which have mor- et al., 2013) as a standalone executable application. The exe- phologically different hypocotyls as shown in Gonzalez-Garcia cutable application together with the datasets generated dur- et al. (2011). ing the current study (from Figures 3, 5 and S4) are available © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156 MyROOT software for plant root length measurement 1155 in the [Zenodo] repository, [https://doi.org/10.5281/zenodo. van den Berg, C., Willemsen, V., Hendriks, G., Weisbeek, P. and Scheres, B. 2552250]. (1997) Short-range control of cell differentiation in the Arabidopsis root meristem. Nature, 390, 287–289. ACKNOWLEDGMENTS Cai, J., Zeng, Z., Connor, J.N., Huang, C.Y., Melino, V., Kumar, P. and Miklavcic, S.J. (2015) RootGraph: a graphic optimization tool for auto- We would like to thank Can~ o-Delgado Laboratory members for mated image analysis of plant roots. J. Exp. Bot. 66, 6551–6562. helping with manual root length measurements and comments on Can~ o-Delgado, A.I., Metzlaff, K. and Bevan, M.W. (2000) The eli1 mutation the manuscript. AIC-D is a recipient of a BIO2016-78150-P grant reveals a link between cell expansion and secondary cell wall formation funded by the Spanish Ministry of Economy and Competitiveness in Arabidopsis thaliana. Development, 127, 3395–3405. and Agencia Estatal de Investigacio  n (MINECO/AEI) and Fondo Clark, R.T., Famoso, A.N., Zhao, K., Shaff, J.E., Craft, E.J., Bustamante, C.D., Europeo de Desarrollo Regional (FEDER), and a European McCouch, S.R., Aneshansley, D.J. and Kochian, L.V. (2013) High-through- put two-dimensional root system phenotyping platform facilitates genetic Research Council, ERC Consolidator Grant (ERC-2015-CoG – analysis of root growth and development. Plant, Cell Environ. 36, 454–466. 683163). IB-P is funded by the FPU15/02822 grant from the Span- Dalal, N. and Triggs, B. (2005) Histograms of oriented gradients for human ish Ministry of Education, Culture and Sport. DB-E is funded by detection. IEEE Computer Society Conference on Computer Vision and the ERC-2015-CoG – 683163 granted to the AIC-D laboratory. This Pattern Recognition, 1, 886–893. project has received funding from the European Research Council Dolan, L., Janmaat, K., Willemsen, V., Linstead, P., Poethig, S., Roberts, K. (ERC) under the European Union’s Horizon 2020 Research and and Scheres, B. (1993) Cellular organisation of the Arabidopsis thaliana Innovation Programme (Grant Agreement No. 683163). This work root. Development, 119, 71–84. was supported by the CERCA Programme from the Generalitat de Fabregas, N., Lozano-Elena, F., Blasco- Escamez , D. et al. (2018) Overex- Catalunya. We acknowledge financial support from the Spanish pression of the vascular brassinosteroid receptor BRL3 confers drought resistance without penalizing plant growth. Nat. Commun. 9, 4680. Ministry of Economy and Competitiveness (MINECO), through the French, A., Ubeda-Tomas, S., Holman, T.J., Bennett, M.J. and Pridmore, T. ‘Severo Ochoa Programme for Centres of Excellence in R&D’ (2009) High-throughput quantification of root growth using a novel 2016–2019 (SEV-2015-0533)’. image-analysis tool. Plant Physiol. 150, 1784–1795. Glumov, N.I., Kolomiyetz, E.I. and Sergeyev, V.V. (1995) Detection of CONFLICT OF INTEREST objects on the image using a sliding window mode. Opt. Laser Technol. 27, 241–249. The authors declare no conflicts of interest. Gonzalez-Garcia, M.P., Vilarrasa-Blasi, J., Zhiponova, M., Divol, F., Mora- Garcia, S., Russinova, E. and Can ~ o-Delgado, A.I. (2011) Brassinosteroids AUTHOR CONTRIBUTIONS control meristem size by promoting cell cycle progression in Arabidopsis roots. Development, 138, 849–859. AIC-D conceived the idea. AG and XS developed the algo- Gonzalez-Garcia, M.P., Pavelescu, I., Canela, A., Sevillano, X., Leehy, K.A., rithms for the method. AG, XS, IB-P and DB-E performed Nelson, A.D., Ibanes, M., Shippen, D.E., Blasco, M.A. and Can~ o-Delgado, the validation experiments. IB-P and DB-E acquired the A.I. (2015) Single-cell telomere-length quantification couples telomere length to meristem activity and stem cell development in Arabidopsis. dataset. XS and AIC-D designed and supervised the study. Cell Rep. 11, 977–989. IB-P, AG, XS and AIC-D wrote the manuscript. Hauser, M.T., Morikami, A. and Benfey, P.N. (1995) Conditional root expan- sion mutants of Arabidopsis. Development, 121, 1237–1252. SUPPORTING INFORMATION Ishikawa, H. and Evans, M.L. (1995) Specialized zones of development in roots. Plant Physiol. 109, 725–727. Additional Supporting Information may be found in the online Iyer-Pascuzzi, A., Simpson, J., Herrera-Estrella, L. and Benfey, P.N. (2009) version of this article. Functional genomics of root growth and development in Arabidopsis. Figure S1. Laboratory setup for taking pictures of the plates. Curr. Opin. Plant Biol. 12, 165–171. Figure S2. The ruler identification process. J€urgens, G., Mayer, U., Busch, M., Lukowitz, W. and Laux, T. (1995) Pattern formation in the Arabidopsis embryo: a genetic perspective. Philos. Figure S3. Evaluation of the time required to measure root length. Trans. R. Soc. Lond. B Biol. Sci. 350, 7. Figure S4. Validation of MyROOT batch analysis processing. Kuijken, R.C., van Eeuwijk, F.A., Marcelis, L.F. and Bouwmeester, H.J. Figure S5. Comparison of MyROOT, BRAT and EZ-Rhizo soft- (2015) Root phenotyping: from component trait in the lab to breeding. J. wares. Exp. Bot. 66, 5389–5401. Table S1. Comparison of available semi-automatic softwares for Le Bot, J., Serra, V., Fabre, J., Draye, X., Adamowicz, S. and Pages, L. quantification of root traits. (2010) DART: a software to analyse root system architecture and devel- Table S2. Comparison of MyROOT, BRAT and EZ-Rhizo softwares. opment from captured images. Plant Soil, 326, 13. Li, J., Nam, K.H., Vafeados, D. and Chory, J. (2001) BIN2, a new brassinos- Video S1. Demo video of MyROOT software. teroid-insensitive locus in Arabidopsis. Plant Physiol. 127, 14–22. Lobet, G. (2017) Image analysis in plant sciences: publish then perish. Trends Plant Sci. 22, 559–566. REFERENCES Lobet, G., Pages, L. and Draye, X. (2011) A novel image-analysis toolbox enabling quantitative analysis of root system architecture. Plant Physiol. 157, 29–39. Armengaud, P., Zambaux, K., Hills, A., Sulpice, R., Pattison, R.J., Blatt, M.R. Lobet, G., Draye, X. and Perilleux, C. (2013) An online database for plant and Amtmann, A. (2009) EZ-Rhizo: integrated software for the fast and image analysis software tools. Plant Methods, 9, 38. accurate measurement of root system architecture. Plant J. 57, 945–956. Lobet, G., Pound, M.P., Diener, J. et al. (2015) Root system markup lan- Arsenault, J., Poulcur, S., Messier, C. and Guay, R. (1995) Winrhizo: a root guage: toward a unified root architecture description language. Plant measuring system with a unique overlap correction method. HortS- Physiol. 167, 617–627. cience, 30, 906. Mouchel, C.F., Briggs, G.C. and Hardtke, C.S. (2004) Natural genetic varia- Beemster, G.T. and Baskin, T.I. (1998) Analysis of cell division and elonga- tion in Arabidopsis identifies BREVIS RADIX, a novel regulator of cell tion underlying the developmental acceleration of root growth in Ara- proliferation and elongation in the root. Genes Dev. 18, 700–714. bidopsis thaliana. Plant Physiol. 116, 1515–1526. Nagel, K.A., Putz, A., Gilmer, F. et al. (2012) GROWSCREEN-Rhizo is a novel Benfey, P.N., Linstead, P.J., Roberts, K., Schiefelbein, J.W., Hauser, M.T. phenotyping robot enabling simultaneous measurements of root and and Aeschbacher, R.A. (1993) Root development in Arabidopsis: four shoot growth for plants grown in soil-filled rhizotrons. Funct. Plant Biol. mutants with dramatically altered root morphogenesis. Development, 39, 891–904. 119, 57–70. © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156  n-Putze et al. 1156 Betego Otsu, N. (1979) A threshold selection method from gray-level histograms. the brevis radix mutant partly reflects root abscisic acid hypersensitivity. IEEE Trans. Syst. Man Cybern. 9, 5. Plant Physiol. 149, 1917–1928. Pace, J., Lee, N., Naik, H.S., Ganapathysubramanian, B. and Lubberstedt, T. Slovak, R., Goschl, C., Su, X., Shimotani, K., Shiina, T. and Busch, W. (2014) (2014) Analysis of maize (Zea mays L.) seedling roots with the high- A scalable open-source pipeline for large-scale root phenotyping of Ara- throughput image analysis tool ARIA (Automatic Root Image Analysis). bidopsis. Plant Cell, 26, 2390–2403. PLoS ONE 9, e108255. Takatsuka, H. and Umeda, M. (2014) Hormonal control of cell division and Pfister, A., Barberon, M., Alassimone, J. et al. (2014) A receptor-like kinase elongation along differentiation trajectories in roots. J. Exp. Bot. 65, mutant with absent endodermal diffusion barrier displays selective nutri- 2633–2643. ent homeostasis defects. ELife 3, e03115. Ubeda-Tomas, S., Swarup, R., Coates, J., Swarup, K., Laplaze, L., Beemster, Potuschak, T., Lechner, E., Parmentier, Y., Yanagisawa, S., Grava, S., Koncz, G.T., Hedden, P., Bhalerao, R. and Bennett, M.J. (2008) Root growth in C. and Genschik, P. (2003) EIN3-dependent regulation of plant ethylene Arabidopsis requires gibberellin/DELLA signalling in the endodermis. hormone signaling by two Arabidopsis F box proteins: EBF1 and EBF2. Nat. Cell Biol. 10, 625–628. Cell, 115, 679–689. Verbelen, J.P., De Cnodder, T., Le, J., Vissenberg, K. and Baluska, F. (2006) Pound, M.P., French, A.P., Atkinson, J.A., Wells, D.M., Bennett, M.J. and The root apex of Arabidopsis thaliana consists of four distinct zones of Pridmore, T. (2013) RootNav: navigating images of complex root archi- growth activities: meristematic zone, transition zone, fast elongation tectures. Plant Physiol. 162, 1802–1814. zone and growth terminating zone. Plant Signal. Behav. 1, 296–304. Ristova, D., Rosas, U., Krouk, G., Ruffel, S., Birnbaum, K.D. and Coruzzi, Wu, J., Wu, Q., Pages, L., Yuan, Y., Zhang, X., Du, M., Tian, X. and Li, Z. G.M. (2013) RootScape: a landmark-based system for rapid screening of (2018) RhizoChamber-Monitor: a robotic platform and software enabling root architecture in Arabidopsis. Plant Physiol. 161, 1086–1096. characterization of root growth. Plant Methods, 14, 44. Rodrigues, A., Santiago, J., Rubio, S., Saez, A., Osmont, K.S., Gadea, J., Yazdanbakhsh, N. and Fisahn, J. (2012) High-throughput phenotyping of Hardtke, C.S. and Rodriguez, P.L. (2009) The short-rooted phenotype of root growth dynamics. Methods Mol. Biol. 918, 21–40. © 2019 The Authors. The Plant Journal published by John Wiley & Sons Ltd and Society for Experimental Biology., The Plant Journal, (2019), 98, 1145–1156