Arabidopsis thaliana (accession Col-0) was used in all experiments. Seeds were surface-sterilized, sown on square plates (12 cm × 12 cm) containing 0.5× Murashige-Skoog (MS) medium (Murashige and Skoog, 1962) (pH 5.7) supplemented with a gelling agent 0.6% Phytagel (Sigma–Aldrich, Germany) and maintained for 3 days at 4°C in the dark. Thereafter, the plates were transferred into a growth-chamber with controlled conditions (22°C, 16/8 h light/dark cycle, a photon irradiance of 120 μmol photons of PAR m^-2 s^-1) and placed in a vertical position. Three days after germination, seedlings of similar size were transferred under sterile conditions into the multi-well plates [12- and 24-well plates (Jetbiofil, Guangzhou, China)] one seedling per well and the plates were sealed with perforated transparent foil allowing gas and water exchange. Each well contained 2.7 mL (12-well plate) or 1.3 mL (24-well plate) of full MS medium (pH 5.7; supplemented with 0.6% Phytagel). For optimization, different concentrations of MS (1×, 0.5×, and 0.25×) and sucrose (0, 0.1, and 1%) (pH 5.7; containing 0.6% Phytagel) were also used. In the salt-stress experiment 12- and 24-well plates were used filled with 1× MS medium (pH 5.7; containing 0.6% Phytagel) with the addition of NaCl to achieve specific salinities (50, 75, 100, and 150 mM NaCl). In the experiment dealing with interacting growth conditions, 12-well plates containing different MS concentrations (1×, 0.5×, and 0.25×) with or without salt stress (75 mM NaCl) were used.

The multi-well plates with the transferred Arabidopsis seedlings were placed onto the OloPhen platform¹ that uses the PlantScreen^TM XYZ system installed in a growth chamber with a controlled environment and cool-white LED and far-red LED lighting (Photon Systems Instruments, Brno, Czech Republic). The conditions were set to simulate a long day with a regime of at 22°C/20°C in a 16/8 h light/dark cycle, an irradiance of 120 μmol photons of PAR m^-2 s^-1 and a relative humidity of 60%. The PlantScreen^TM XYZ system consists of a robotically driven arm holding an RGB camera with customized lighting panel and growing tables with a total area of approximately 7 m² with a capacity of 480 multi-well plates fixed in customized trays for accurate positioning of every plate (Figure 1A). The XYZ robotic arm was automatically moved above the plates to take RGB images of single plates from the top. RGB images (resolution 2500 × 2000 pixels) of a single plate with a file size of approximately 10 MB in the PNG compression format (Figure 1B) were stored in a database on a server, using a filename containing information about the acquisition time and the (x, y) coordinates of the camera. The data were automatically stored in PlantScreen XYZ database, exported by PlantScreen Data Analyzer software and analyzed using an in-house software routine implemented in MatLab R2015 (for details see Results). The application can be used without any charge upon obtaining a license from the author. The license can be obtained by e-mail to Palacky University upon agreeing not to use the application for commercial purpose. After obtaining the license, the end-user will be provided (free of charge) with the MCRInstaller.exe. MCRInstaller simulates the MatLab environment on computers where MatLab is not installed and enables to execute the applications. To obtain the application executable files, please contact the author Tomas Furst by email tomas.furst@upol.cz. The email must contain the following statement: “Neither the application nor the MCRInstaller will be used for any commercial purpose”.

To assess the differences between the projected areas of two or more groups of plants at a particular time-point, the non-parametric Kruskal and Wallis one-way analysis of variance by ranks was used. The test compares the medians of the samples in the respective groups, and returns a p-value for the null hypothesis that all samples are drawn from the same population. For the analysis of multidimensional data, visual representations in the form of box plots, histograms, and animations, created using the MatLab R2015 software, were used to capture the time dimension of the problem.

The presented assay was designed for automated large-scale analysis of Arabidopsis rosette growth in vitro using multi-well plates. The in vitro cultivation of the plants confers the advantage of precise control of the growth media and easy supplementing and dosing of tested factors. Importantly, it allows easy introduction of generalized randomized block designs (GRBDs), the statistical theory of the design of experiments that is used to study the interaction between blocks and treatments. In our method, a block is represented by a multi-well plate containing an array of plants. The plates can then be randomized within the growth area and replicated in the case of optimization of the method when a blocking factor is tested as the potential source of variability. Typically, such a source is differences in the growth chamber microclimate, however, one usually underestimated factor that introduces nuisance variables is the operator preparing the treatment.

Thus, to ensure an appropriate experimental procedure, we established the protocol schematized in Figure 2; this was used for optimization of the method, its validation, salt stress response and growth interaction studies. The protocol takes a total of 18 days and consists of several steps including seed sowing, cold stratification, transferring the seedlings into the multi-well plates, time-course RGB-imaging, data processing and analysis (Figure 2). The seeds are sown onto the square plates (12 cm × 12 cm) containing compound/stressor-free medium at a density of about one seed per 1.5 cm², facilitating manipulation during subsequent seedling transfer. After cold stratification, the plates are placed vertically in the growth chamber, thus preventing growth of the root into the solid media. This minimizes any possible damage to the tiny 3 days after germination (DAG) old seedlings during the transfer into the multi-well plates (one seedling per well). The transfer of the seedlings was introduced into the protocol as an important step to achieve a method in which the effect of the tested conditions (stressors, chemicals, etc.) is scored not earlier than during the stage when the cotyledons are expanded, thus avoiding the possible effect on the process of germination. As described later in the text, the selection of seedlings of similar size for the transfer into multi-well plates represents a critical point in this method. After seedling transfer, the multi-well plates are placed into the PlantScreen^TM XYZ system and after about 24 h of acclimation, automated RGB-imaging is performed every day for the next 9 days. When the platform is at capacity, the whole imaging run takes 70 min, thus in theory there can be 12–13 imaging runs within the 16-h-long light period, producing a dense-point growth curve. In the case of a well optimized assay, the sensitivity of the analysis allows scoring differences in the rosette area over only 2-h-long intervals. As shown in the example of two independent 12-well plates, the average increase in the green area in 2 h is 2-3% (Supplementary Figure S1). This offers the possibility for further optimization to increase the assay through reduction of the entire time of the assay and/or use of multi-well plates with a larger number of smaller wells. In the presented protocol, the growth of the green area was recorded every 24 h (typically at midday) during the 9 days. The imaging data were processed using in-house software described in the following section.

The RGB imaging data were analyzed using an in-house software routine, examining all the files covering the entire experiment, i.e., images of up to 1000s of plates taken at many pre-defined time-points. In the first step, fish-eye correction of each image is performed. Next, the image is registered automatically so that the positions of the wells are correctly identified. For this step, blue boundary marks on the trays are used (Figure 1C), together with an edge detection routine. The registration step is somewhat sensitive to errors, therefore several suggestions for the correct registration are computed and returned in order of decreasing probability. The most probable registration is tried first and the image is registered and cropped to contain only the plate with no surroundings. The plants are automatically detected by thresholding the image in the HSV color space. Since we are looking for green pixels in a generally white background, the threshold need not be very intelligent, a fixed cut-off value is used separately in each of the three HSV channels and the results are combined by means of the logical operator “AND.” If any green area is detected outside the wells, the software recognizes that it made a mistake in the registration step and returns to try the registration step with another set of plausible registration parameters. When the segmentation is successful, i.e., there is no green area detected outside the wells, the green areas of all the plants are computed and a pre-view of the registered and segmented image is saved to disk. The pre-views can be reviewed manually and any remaining errors corrected. After all the images have been analyzed, a single XLSX data file is produced which contains, in each row: the name of the file, date of acquisition, (x, y) position of the camera, and subsequently the list of 12, or 24 numbers, which represent the green areas (in pixels) of the plants in the wells. The wells are numbered column-wise. On a standard PC, the processing of a single 10 MB PNG file takes approximately 10 s with most of the time spent on the fish-eye correction routine. Since the experiment itself usually lasts for days, there was no need for any speed-optimization of the MatLab routine.

For correct data handling, we must take into account the fact that the data produced by these experiments are naturally multi-dimensional. Usually, several different treatments in various concentrations are tested together with one or more controls, with the option to include different numbers of wells per plate. Thus, there are at least three independent predictors of the green area: time, type of treatment (together with its concentration), and the type of plate (6-, 12-, or 24-well). It is also important to keep track of plants that come from the same plate because it is possible that there is more correlation between the green areas of plants from the same plate than among plants from different plates. These multidimensional data are not easy to handle using standard table-processing software (e.g., Excel) because of a richer data-structure, so a table with more than two dimensions is needed. Thus, the data processing was also performed using MatLab software, creating a data structure with the same length as the number of PNG files. Each item in the structure contains the following terms: time from the beginning of the experiment, type of treatment, numerical code of the treatment, type of plate, position on the table, and a list of the green areas of the plants. The same data structure was used for the statistical analysis and data representation.

The right experimental design for the assay requires introduction of a standardized protocol resulting in maximum homogeneity of the recorded trait, in this case similar plant growth on a single plate and among plate replicates. This allows application of the statistical methods to describe significance of the differences between the tested variants. To define the most suitable screening conditions to achieve HTS using in vitro conditions, first the response of 4-day-old Arabidopsis seedlings grown in 1× MS [recommended for Arabidopsis growth in the protocol published by Cold Spring Harbor Protocols (Recipe, 2010) and by the Arabidopsis Biological Resource Center²] was evaluated using a different culture plate format, with a higher number of replicates randomly distributed across the growth area. The 12- and 24-well plates were prepared following the experimental scheme (Figure 2) with nine and six replicates per variant (represented by a single plate), respectively, and the rosette size (represented by the green area) was analyzed for 9 days. The outcome of the analysis can be either a single growth curve describing the increase in green area over time (Figure 3A), or a curve showing the relative growth rate calculated as described Hoffmann and Poorter (2002) (Figure 3B). The curves for seedling growth in the 12-well and 24-well plates had similar profiles, showing that during the 9 days when data were collected, the green area of the seedling exhibits high significant exponential growth (Figure 3A). Although the rosette area of the seedlings grown in the 12- and 24-well plates starts to differ after 1 week of cultivation, the relative grow rate (RGR) of the seedlings shows the same tendency, with parallel curves but with higher values for the those grown in 12-well plates (Figure 3B). This indicates that the volume and space of the well are the main factors determining the difference observed after the sixth day.

To optimize the assay we used a Kruskal–Wallis test to evaluate statistically the differences in rosette area between the plate replicates, which were randomly distributed within the growth area. Unexpectedly, significant differences in the average green areas, in some cases reaching almost 50%, were observed within the plate replicates in both types of multi-well plate at the analyzed time-points (Figure 4). Analysis of the rosette growth in the replicates with the smallest, intermediate and largest average growth areas (R3, R4 and R8 or R1, R2 and R5 for 12- and 24-well plates respectively) revealed similar profiles (Figures 3C,D), with significant differences between the two extremes (Table 1). Taking into account the experimental set-up of the assay, the possible reasons for the differences in the average rosette size in the randomly distributed replicates could be either different micro-climatic conditions in the growth chamber or the non-randomized selection of the seedlings at the time of transfer with respect to developmental stage, resulting from the natural heterogeneity of the population. Measurements of the micro-meteorological conditions in the phenotyping chamber did not reveal any differences. For this reason, in the next step we increased the number of germination plates to increase the population of the seedlings. This allowed us to improve our selection of the 4-day-old seedlings, ensuring that they were all a similar size at the time of transfer, discarding any particularly large or particularly small seedlings. The more careful selection of seedlings did, indeed, result in standardization of population heterogeneity between plate replicates, and almost no statistically significant differences between the average rosette areas of the nine replicates were found at the different time-points (Figure 5).

Finally, to test the reliability of the method, we compared the green area estimated by automated RGB imaging with the weight of the rosettes determined manually. The rosette of individual plants grown in vitro in 24-well plates containing 1× MS and 0.5× MS medium, respectively, were harvested on the last day of measurements and the fresh-weight (FW) of individual plant rosettes was determined. Subsequently, correlations between the green area and FW were calculated using Pearson’s coefficient and the significance determined after ANOVA. In both growing conditions, a highly significant correlation was obtained with correlation coefficients of 0.94 and 0.85, respectively (Figure 6). The relationship between green area and FW of Arabidopsis rosettes showed more homogeneous size of the plants grown in 0.5× MS compared to 1× MS. In 1× MS conditions the distribution of the population was broader, with higher number of smaller (<10 mg of FW) and bigger plants (>40 mg of FW), suggesting that the nutritional conditions contribute to the population phenotype.

To select our standardized normal conditions we tested experimentally whether MS concentration influenced Arabidopsis rosette growth and also evaluated the need for sucrose as a component of the growing medium. First, MS medium without sucrose was used in different concentrations; 0.25×, 0.5×, and 1×. A clear concentration-dependent increase in rosette area was found, indicating that 1× MS is the best growing medium for Arabidopsis seedlings in vitro (Figure 8A). Although the RGR of the seedlings grown on 0.5× MS was comparable to those on 1× MS during the first 4 days, in the second half of the growth period the development of these seedlings slowed and their RGR was 13% lower than for seedlings on 1× MS, decreasing to 25% during the last 2 days (Figure 8B). This suggests that use of lower MS concentration than 1× can result in a change from optimal to suboptimal growth conditions during the period that the experiment is running and the seedlings are inadvertently subjected to low nutrient stress during their, otherwise, exponential growth period.

As mentioned above, the use of sucrose in growth media is also generally very variable, ranging from a concentration of 0 to 3%. As discussed later, the presence of sucrose in the growing medium leads to substantial changes in the physiology of the developing seedling, conditioning seed germination and modifying plant metabolism (Ohto et al., 2001; Eckstein et al., 2012). Hence, we tested how the exogenous addition of sucrose alters the growth of Arabidopsis seedlings in optimal nutritional conditions and compared plant performance when grown on 1× MS medium containing 0, 0.1, and 1% sucrose. No significant differences were found in the increase of the rosette areas of the seedlings grown with and without sucrose over the duration of the experiment (Supplementary Figure S2). Taking into account these results, we decided to use 1× MS without sucrose as the standard growing medium for our assay.

Our platform has sufficient capacity to allow simultaneous testing of large numbers of variants. This can be employed for evaluation of chemical libraries and/or genetic populations in normal and stressed conditions and for cross-testing of a wide range of concentrations of stressors and/or tested compounds. To illustrate the potential of our assay to be used as a tool for large-scale stress-response studies, we performed an experiment in which the effect of salt on Arabidopsis rosette growth was tested using 1× MS medium supplemented with different concentrations of NaCl (50, 75, 100, and 150 mM). Three replicates of a 24-well plate were used for each tested variant, with no significant differences among them throughout the experiment (Figure 9). Both time-dependent increase in shoot area and RGR were found to be negatively affected by NaCl treatment in a dose-dependent manner (Figures 10A,B). Even after 2 days, significant differences in the rosette area were recorded between the controls and the plants grown in the presence of 100 mM and 150 mM NaCl (Figure 10A and Table 2), due to a very fast decrease in RGR: 56 and 84%, respectively (Figure 10B). After 5 days, significant differences in the rosette area were also apparent between the controls and the lowest salt treatments (Table 2). Interestingly, salt treatment modified the population distribution causing changes in the quartiles. The moderate salinity (50 mM) increased the rosette areas of the plants of the first and the third quartiles (Q1, Q3) until the fifth day, after which the salinity started to have the expected negative effect on rosette growth (Table 2). The severe salt-stress conditions (100 mM and 150 mM NaCl) had clear negative effects on the rosette growth and, moreover, reduced the plant size in both quartiles. The Q1 for the plants treated with 150 mM NaCl was reduced to zero on the fifth day of the salt treatment and the survival of plants in this variant reduced from 67% at day 7 to 50% at day 9 (Figure 10C). Overall these results proved the potential of the assay to be used as a tool for salt-stress studies.

In the previous text we described the effect of different salt concentrations on the Arabidopsis rosette grown under optimal nutrient conditions (1× MS). As mentioned above, analysis of the typical conditions used in salt-stress studies, 0.5× MS was mostly chosen as the source of nutrients (Figure 7C). This fact led us to perform an experiment in which the effect of 75 mM NaCl (identified in this study as representing medium salt stress) on Arabidopsis rosette growth was tested in MS medium of three different strengths, i.e., 0.25×, 0.5×, and 1×. Each variant comprised three replicates on a 12-well plate and no statistical differences among them were found according to a Kruskal–Wallis test (Supplementary Figure S3). When the green area of the different treatments was analyzed we obtained an unexpected result: a significant interaction between MS concentration and salt treatment (Table 3 and Figures 11A,B). Both time-dependent increase in the green area of the rosette and decrease in RGR were higher in salt-stressed plants grown in 0.5× MS than those in 1× MS (Figures 11A,B). When they were compared with the plants grown in different MS without salt (Figure 8A), we observed that whereas plants grown in 1× MS without salt had at least 2-4 times bigger rosettes compared to the salt stressed ones after 7 and 9 days of the treatment, respectively, no significant differences were observed between those Arabidopsis grown with and without 75 mM NaCl in 0.25× and 0.5× MS (Figures 11C,D and Table 3). Interestingly, whereas the salt treatment reduced the quartiles, median and average rosette area of the plants grown in 1× MS by a factor of four, whilst keeping similar minimum and maximum values, it improved the Q1 and Q3, and the minimum size of the plants grown in the low nutrient media (Table 3). These results were further confirmed by the population distribution of each treatment over time (Supplementary Figure S4), where the salt treated 0.25× and 0.5× MS variants presented a narrower distribution and more homogeneous populations with plants of similar rosette size compared to their respective controls. The analysis also revealed the same average green areas, size heterogeneities and similar distribution of the populations of the plants grown in 0.25× MS and plants grown in 1× MS with 75 mM NaCl (Table 3, Figures 8, 11, and Supplementary Figure S4). These results revealed the existence of a crucial interaction between the concentration of nutrients and the salt treatment that conditions the stress response and growth capacity of the plants through the heterogeneity of the plant population.