MADI: A multispectral automated dynamic imager to monitor plant health

Abstract

High-throughput phenotyping has a tremendous capacity to advance our understanding of plant biology. Integrating growth parameters with information on a plant's physiology through multispectral imaging can provide a holistic picture of its health status and its responses to environmental stressors. Furthermore, the screening of large-scale populations of genotypes or germplasms, using such platforms, can identify lines with desirable traits to help feed a growing world population in the background of climate change. Here, we present a novel platform, the Multispectral Automated Dynamic Imager (MADI), which combines visible and near-infrared reflectance, thermal imaging, and chlorophyll fluorescence for the dynamic monitoring of growth, leaf temperature, and photosynthetic efficiency. Additionally, we have integrated and validated a fluorescence-based parameter to non-destructively assess chlorophyll content. The utility of the MADI system was demonstrated through four case studies in which lettuce and Arabidopsis plants were exposed to various abiotic stress conditions. We demonstrate that plant compactness is a useful marker for stress responses, including drought, and could serve as a biomarker to study plant hormones. Additionally, we observed the phenomenon of chlorophyll hormesis under salt stress, a rather poorly understood process. In conclusion, the MADI is a multifunctional, adaptable system that can be employed to gain insights into plant stress responses and help to improve agricultural practices. It can be used primarily for rosette-growing species, such as leafy greens, which represent a significant portion of cultivated crops worldwide.

1. Introduction

Plant development and health are under constant pressure of a changing environment. In the current climate emergency, demands on agricultural production are rising. Moreover, they are worsened by the increasing need for food caused by a growing world population [1] and by the threat of declining macro- and micronutrient content of crop products [2]. Therefore, tools to assess and control plant growth and health are crucial for future agriculture. In addition, efforts are being made to identify useful cultivars, either by using genetic variation present in natural populations, or through the generation of transgenic crops with desirable traits [3]. Though high-throughput technologies for generating and mapping genomics data have been developed, the capacity to link those genetic changes with desirable traits and phenotypes still is a bottleneck [4]. Therefore, the current challenges for highly efficient agri- and horticultural practices are driving the advances in plant monitoring systems. High-throughput phenotyping (HTP) technologies for non-invasive monitoring of plant functioning, including growth parameters as well as physiological status, are being developed for both fundamental and applied research purposes, and within the field of precision agriculture [[5], [6], [7]]. In this context, visualization of a large number of plants with a high spatial and temporal resolution is desired. These phenotyping platforms can be field-based (e.g. Phenobot [8]) or housed in controlled environments such as greenhouses and laboratories (e.g. PHENOPSIS [9], Phenoscope [10], PHENOARCH [11], DEPI platform [12]), and can integrate growth-based parameters as well as other physiological proxies. Each of these platforms possesses distinct strengths and limitations, which stem from their specific imaging approach [e.g., plant-to-sensor versus sensor-to-plant systems), imaging capabilities (e.g., the number of plants, duration of imaging, plant architecture-related (e.g., overhead or lateral imaging)], potential for environmental control (e.g., temperature, soil humidity, CO₂ level), and the range of parameters they can extract (e.g., growth, water status, photosynthesis, temperature). In recent years, phenotyping robots for farming applications also found their way to the market, as for instance the CLAWS robot (Earth Rover, Newport, UK) and Vulcan (FarmWise, CA, USA).

Automated imaging frequently consists of measurement and estimation of growth parameters [13] and is included in a variety of available phenotyping platforms. These parameters differ according to plant architecture and include, among others, length, biomass, and area [14]. However, more information can be gathered combining such analyses with direct or indirect monitoring of physiological parameters. An indirect measurement of stomatal conductance and transpiration can, for example, be obtained by thermal imaging [5]. Consequently, leaf temperature differences read by thermography have been employed to isolate mutants with altered stomatal conductance [15,16]. Additionally, thermal imaging can be used to visualize differences in irrigation intensity and for the early detection of pathogen infections [[17], [18], [19], [20], [21]]. Chlorophyll fluorescence is also frequently used to estimate if a plant is under stress [22,23]. The most widely applied parameters are the maximum quantum efficiency of photosystem II (PSII) (or F_v/F_m) and the operating efficiency of PSII (F_q’/F_m’ or ΦPSII), both estimating the efficiency at which light absorbed by PSII is employed to reduce quinone A [22]. The former parameter differs from the latter in that F_v/F_m is measured in dark-adapted plants, while F_q’/F_m’ in light-adapted plants. The use of handheld devices that can estimate these parameters in single leaves is a common method to track plant physiology, especially in field experiments. Nevertheless, the adaptation of this method to photographs allows the analysis of multiple leaves and/or multiple individual plants at the same time.

Chlorophyll content, on the other hand, relates directly to the potential for photosynthesis and growth of plants [24]. Changes in chlorophyll content or distribution are associated with a plethora of physiological conditions. These range from nutrient deficiencies (e.g. nitrogen, magnesium, sulfur, among others) to senescence and (a)biotic stress [24,25]. It is therefore of particular interest to estimate chlorophyll content in plant leaves. Total chlorophyll can be destructively measured after extraction through spectrophotometric measurements of the pigments [26]. Non-destructive methods for chlorophyll estimation and plant detection vary widely. For imaging, a well-known and used indicator for leaf greenness is the normalized difference vegetation index NDVI [27]. While chlorophyll fluorescence has been used to probe photosynthesis [22] and to detect stress [19,28,29], it was also proposed that the ratio of fluorescence at 735 ​nm versus 700 ​nm (F₇₃₅/F₇₀₀) could be an accurate measure of chlorophyll content in plants [30]. Nevertheless, this estimation has not been implemented in HPT known to date.

Here we describe an in house built Multispectral Automated Dynamic Imager (MADI) sensor-to-plant phenotyping platform that combines robotized imaging based on visible and near-infrared reflectance (NIR), thermal imaging, and chlorophyll fluorescence to extract morphological and physiological data from plants (Fig. 1). We show that, using this system, it is possible to predict drought stress based on visible, fluorescence or thermal images. Moreover, our results highlight the utility of compactness as a marker for plant stress. In addition, a parameter for the quantification of chlorophyll concentration in plants was tested, proven accurate, and incorporated into the MADI protocol. The latter was validated for the dynamic evaluation of chlorophyll levels over time and uncovered the existence of chlorophyll hormesis during salt stress. The proposed parameter can be applied in prediction and early detection of external stressors and plant senescence. The system has been successfully tested on lettuce and Arabidopsis and can hold up to over 60 and 600 plants of the respective species. Compared to similar HTP platforms in controlled environments, the MADI excels because of minimal plant manipulation, a diverse set of imaging capabilities, including chlorophyll content determination and an optimized chlorophyll fluorescence protocol, and enhanced modularity of the system due to the inclusion of a filter wheel for different bandpass filters.

Fig. 1.

Components of the MADI system. (A) Ebb-and-flow cultivation table containing trays for growing plants. (B) Ebb-and-flow tank. (C) MADI imaging box housing the VIS-NIR camera, filter wheel, focus gears and LEDs. (D) Long-wave infrared camera for thermal imaging. The thermal camera is attached to the outside of the MADI imaging box.

2. Materials and methods

2.1. Experimental design of the MADI system

The MADI system is housed in a climate-controlled growth chamber (Weiss Technik, Belgium). It consists of the following modules: 1) a cultivation table to grow plants with an ebb and flow irrigation system, 2) an imaging box containing light emitting diodes (LEDs), two cameras and electronics, mounted on 3) a robotic arm with three axes (xyz), and 4) a dedicated computer, which controls all other modules (Fig. 1). A detailed description of the MADI is given below.

2.1.1. Growth chamber and cultivation table

The growth chamber is equipped with two types of gas discharge lamps. Four full spectrum D-light Papillon High Pressure sodium lamps (315 ​W) (Lux Light International, Marburg, Germany), and four Master Color CDM-TP MW 315 ​W/930 lamps (Philips, Zellik, Belgium). All lamps generate 300 ​μmol ​m⁻² ​s⁻¹ photosynthetic photon flux density (PPFD) at the plant level. A 150 ​× ​250 ​cm cultivation table is positioned at the center of the chamber (Fig. 1A), which is connected to an ebb and flow irrigation system for automatic watering (Fig. 1B). On this table, 16 plant trays (30 ​× ​50 ​cm; Aratray system, Betatech, Ghent, Belgium) can be placed, each of which can fit 51 Arabidopsis thaliana plants. However, only up to 38 plants are useful for imaging. For larger species, the maximum number of plants is lower; for instance, the complete system harbors up to 60 full-grown lettuce plants. Via a solid-state relay, the computer can control the water pump of the ebb and flow irrigation system. An adjustable draining controls the water level during irrigation and recycles the water or nutrient solution to a storage tank placed below the table.

2.1.2. MADI imaging box

The imaging box consists of a plastic white box (30 ​cm width x 40 ​cm length x 61 ​cm height) that houses the cameras, filters, LEDs and circuits needed for multispectral imaging (Fig. 1C). A scientific-grade thermal camera (model A655sc; Teledyne FLIR, Wilsonville, Oregon, USA), equipped with a 24.6 ​mm (25°) lens for imaging of long-wave infrared radiation (LWIR), is mounted on the outside of the imaging box (Fig. 1D). For imaging in the visible and near-infrared (VIS-NIR) spectrum, a CCD camera is used (model Manta G-609B/C; Allied Vision, Stadtroda, Germany). The camera includes a 1″ Sony ICX694 CCD sensor with EXview HAD II technology. The camera has a signal-to-noise ratio at saturation (SNR_max) of 41.52 ​dB and a dynamic range (DR) of 62.08 ​dB. A filter wheel is mounted in between the CCD camera and the lens. The used C-mount lens has a focal length of 16 ​mm and an aperture of f1.4 (model Lensagon C8M1614GSV2; Lensation, Karlsruhe, Germany). The filter wheel (model HSFW; Edmund Optics, Barrington, New Jersey, USA) has eight filter positions (Fig. 2A), of which six hold the filters for imaging in the visible and infrared spectrum (see Table 1 for wavelength specifications). Additionally, a custom-made autofocus system was developed; the lens focus ring is adjusted with a set of gears and a stepper motor (Fig. 2A). Inside the box, an aluminum heat sink carries white, blue, far-red and infrared LEDs (Fig. 2B). Twelve high-power white LEDs are required for the acquisition of RGB images as well as certain chlorophyll fluorescence protocols. In addition, 108 high-power blue LEDs with a peak emission at 447 ​nm and 12 far-red LEDs with peak emission at 730 ​nm are used to support chlorophyll fluorescence protocols. Finally, MADI is equipped with 12 infrared LEDs with peak emission at 950 ​nm (not influencing growth; [31]) for plant imaging in darkness. The LEDs are powered with a custom-made driver, housed within an aluminum box, which is laterally attached to the MADI imaging box (Fig. 2C). The LED drivers allow dimming of the LEDs and synchronizing the on/off times with the camera trigger. This is performed with digital signals generated by the control computer. The spectral properties of the LEDS can be found in Fig. 2D. The MADI box's inner walls are covered with reflective material to achieve illumination with higher intensity and more homogeneous distribution (Fig. 1C). Furthermore, a black satin ribbon is attached to the bottom of the box to prevent light leakage to neighboring plants.

Fig. 2.

The MADI imaging box. (A) VIS-IR camera (Manta) with filter wheel (indicated with red arrow) and autofocus gears (yellow arrow). (B) LEDs attached to the ceiling of the MADI imaging box. (C) Aluminum box housing the LED power supply and custom-made LED drives. (D) Spectrum of LEDs mounted in the imaging box. PFD, photosynthetic flux density.

Table 1.

Filters installed in the filter wheel, with 8 positions.

Filter position	Filter name	Type	Dominant wavelength	Full width at half maximum (FWHM)
1	Red	Dichroic	–	–
2	Green	Dichroic	–	–
3	Blue	Dichroic	–	–
4	PSII chlorophyll fluorescence	Bandpass	675 ​nm	50 ​nm
5	700 ​nm	Bandpass	700 ​nm	10 ​nm
6	730 ​nm	Bandpass	730 ​nm	10 ​nm
7	empty	–	–	–
8	empty	–	–	–

2.1.3. Robotized platform and control-computer interface

A cartesian coordinate robot is powered from a single electric cabinet with a dedicated power line for reliable operation (Fig. 1A–C). Moreover, this cabinet also houses motor drivers and controllers. A dedicated desktop computer controls the cameras, LED drivers, robotic arm, as well as the ebb and flow irrigation pump. Both cameras are connected to the computer with a dedicated frame grabber. This frame grabber follows the Gigabit Ethernet (GigE) standard and is equipped with a Power over Ethernet (PoE) functionality. A multifunctional Data Acquisition Device (DAQ) (model PCIe6321; NI, Austin, Texas USA) controls the LED drivers, lens autofocus stepper motor, camera triggers and watering pump solenoid. A custom software developed in LabVIEW (National Instruments, Austin, Texas, USA), using Machine Object Oriented Programming (OOP), allows users to remotely control the MADI imaging system. An overview of the MADI system and the general workflow is presented in Fig. 3 and detailed in the following paragraphs.

Fig. 3.

Concept and workflow of the MADI system. (A) Overview of the MADI system and capabilities. (B) General overview of the workflow.

2.1.4. RGB and infrared imaging

The system has four categories of measurement protocols: reflectance, chlorophyll fluorescence, direct fluorescence and longwave infrared. The reflectance measurement protocol produces RGB and night vision images. For RGB images, three consecutive images are acquired with the VIS-IR camera using the RGB dichroic filters (Table 1). Color balancing is performed using color and flat field calibrations. Night vision images are acquired by turning on the 950 ​nm infrared LEDs, with no filter selected on the filter wheel.

2.1.5. Chlorophyll fluorescence imaging

The protocol for measuring chlorophyll fluorescence emitted from PSII is based on the Dynamic Environmental Photosynthetic Imaging (DEPI) system [12,32]. Both F_v/F_m and F_q’/F_m’ can be acquired using the chlorophyll fluorescence bandpass filter (675 ​nm), the VIS-IR camera, and the blue and white LEDs. For acquiring F_v/F_m, the imaged plants must be dark-adapted for at least 20 ​min. Following Baker [22], F_v/F_m estimates and saves the following parameters: minimal fluorescence of a dark-adapted leaf (F₀), maximal fluorescence of a dark-adapted leaf (F_m), the variable fluorescence of a dark-adapted leaf (F_v) which equals F_m- F_o, and F_v/F_m. Several parameters affecting the DEPI protocol can be adjusted, including the time between flashes for F_o and F_m, the measuring beam intensity and saturating pulse intensity, as well as the dark adaptation time. For F_q’/F_m’, the imaged plants are light-adapted. F_q’/F_m’ estimates and saves the following parameters: steady state fluorescence (F′), maximal fluorescence of a light-adapted leaf (F_m’), minimal fluorescence of a light-adapted leaf (F₀’), and F_q’/F_m’. The difference in fluorescence between F_m’ and F’ (=F_q’) is also estimated in the background. To increase signal-to-noise ratio, vertical and horizontal binning is performed.

2.1.6. Chlorophyll content imaging

Estimation of chlorophyll content in fluorescence images was done according to Gitelson et al. [30] with some modifications. First, using blue LEDs as excitation source, fluorescence images at 730 and 700 ​nm are acquired. The ratio between fluorescence emitted at 730 to 700 ​nm (F₇₃₀/F₇₀₀) is further multiplied by 10 thousand to allow the final fluorescence-based F₇₃₀/F₇₀₀ parameter to be stored in standard image file. In addition, F₇₃₀/F₇₀₀ was correlated with experimentally determined chlorophyll content via destructive measurements. Eighty Arabidopsis plants from different developmental stages were used for generating the calibration curves. A linear regression model was fitted and an R² value was calculated (Fig. 4A). Next, higher-order regression models were fitted to the dataset to evaluate the improvement of R² values. The best regression model and equation was kept for the calibration. Similarly, for lettuce (n ​= ​95) and tomato (n ​= ​18) calibration curves were constructed. In all cases, linear regression models best fitted the data. The resulting calibrated correlation is used for non-destructive chlorophyll estimation from F₇₃₀/F₇₀₀ images. The destructive chlorophyll measurements were performed as follows. Leaf discs from Arabidopsis [Col-0 wild type (WT)], lettuce [Lactuca sativa Cecilia R2, (semillas Veyrat, Nunhems)], tomato [Solanum lycopersicum cv moneymaker (Het Vlaams Zaadhuis)], Ctenanthe kummeriana, Calathea makoyana, Maranta leuconeura, Ctenanthe burlemarxii, Calathea lancifolia, Ceratonia siliqua, Quercus suber, Drimys winteri, Fuschia boliviana, and Peumus boldus were obtained using a cork bore. Apart from Arabidopsis, lettuce, and tomato, all species were collected from the ex-situ collection in the Botanical Garden of Ghent University (Ghent, Belgium). Discs were collected in Eppendorf tubes, frozen in liquid nitrogen and pulverized in a ball mill (Retsch, Haan, Germany). Chlorophyll was extracted with 600 ​μl of extraction buffer containing 10 ​% 200 ​mM Tris HCl buffer (pH 8) and 90 % ice cold acetone. Subsequently, the extracts were centrifuged at 4 ​°C at 12000 ​g for 10 ​min. The supernatant was used for spectrophotometric analysis. Absorption at 647 and 664 ​nm were used to determine chlorophyll content according to Ritchie et al. [26]. For proper comparison, values were normalized based on surface area of the discs.

Fig. 4.

Fluorescence by F₇₃₀/F₇₀₀ ratio derived parameter is associated with chlorophyll content. (A) Arabidopsis (Linear regression with β ​= ​0.01733, 95 ​% CI [0.01501, 0.01965], p ​< ​0.0001, R² ​= ​0.7393, n ​= ​80). (B) Lettuce (Linear regression with β ​= ​0.02508, 95 ​% CI [0.02192, 0.02823], p ​< ​0.0001, R² ​= ​0.7280, n ​= ​95). (C) Tomato (Linear regression with β ​= ​0.01897, 95 ​% CI [0.01622, 0.02173], p ​< ​0.0001, R² ​= ​0.9301, n ​= ​18). (D) F₇₃₀/F₇₀₀ ratio in senescent and control plants. Twenty Arabidopsis plants were grown for three weeks in standard growth room conditions. Ten of them were subsequently covered from light for five days (senescent), while ten plants were kept as control. Rosettes were imaged before and after harvesting of a leaf disc of a mature leaf. One leaf disc was taken per plant. The calibration set is derived from the plants depicted in panel A and displayed obvious variation in chlorophyll content/greening.

2.1.7. Thermal imaging

Thermal images were taken with the FLIR camera using an emissivity of 0.96. Emissivity, the ratio of the emitted radiation of the object of interest and a black body is considered to be between 0.93 and 0.98 for plant leaf surfaces [33]. The default value of 0.96 can be adjusted. In order to correct for the camera temperature, a non-uniformity compensation (NUC) calibration is performed before capturing 25 frames at a rate of 50 ​Hz.

2.2. Image (pre-)processing, segmentation, mapping and data-analysis

All images taken by the VIS-IR camera are stored in 16-bit TIFF files. Possible pixel values are integer numbers ranging from 0 to 16384 as the pixel format is set by default to 14-bit resolution. Images resulting from ratios, e.g. F₇₃₀/F₇₀₀ and F_v/F_m, are multiplied by 10000 in order to store decimal information in an image file, which can only store integers. Saved images are 16-bit TIFF files. Possible pixel values are integer numbers ranging from 0 to 65536. The stored RAW pixel values in ratio images must be divided by 10000 to obtain original values. Since the amount of light reaching the VIS-NIR sensor is not homogeneous, as a result of lens and camera vignetting (causing reduced image brightness toward the periphery), a normalized flat field correction is used to obtain correct RGB images and when analyzing chlorophyll fluorescence images of light-adapted plants. Thermal images are saved in a 16-bit TIFF file. Possible pixel values are integer numbers ranging from 0 to 65536. The stored RAW pixel values must by divided by 100 to obtain temperature values in Kelvin. After image acquisition and preprocessing, information is extracted.

Specific regions of interest (ROIs) were selected in ImageJ to obtain average intensity of selected pixels [34]. In addition, RGB images were first used for plant segmentation, as well as for obtaining biometric parameters (projected rosette area, compactness and maximal diameter), segmenting fluorescence and thermal images, and for studying complete plants. Plant segmentation started with identification and elimination of background pixels associated with the plastic tray and soil, which exhibited greyish tones or brown hues, respectively. The approach was based on two hypotheses. First, “greyish” pixels corresponding to the plastic tray have equal intensities across its Red (R), Green (G), and Blue (B) components under an ideal scenario. To account for variation in lighting and image quality, a threshold parameter, θ₁, was introduced to allow small permissible differences between RGB values of pixels pertaining to plastic trays. The hypothesis was defined as follows: for a pixel to be classified as part of the tray, the absolute differences between its color components must remain within the specified threshold θ₁. The threshold can range from 1 to 12, depending on the dataset characteristics. In the datasets analyzed in this work, θ₁ was set at 12 and reliably captures grey tones of the plastic tray. The formal condition for classifying tray pixels is:

TrayPixels_i,j ← | R_i,j ​− ​G_i,j | ​≤ ​θ₁ AND | G_i,j ​− ​B_i,j | ​≤ ​θ₁

This method ensures precise segmentation of tray pixels while minimizing false positives. The second hypothesis aimed at classifying and removing soil and algae pixels, which typically exhibit brownish hues, turning them into background pixels. These pixels consistently had a dominant red component compared to the green and blue channels. Therefore, the second hypothesis was formulated as follows: a pixel is classified as soil or algae if the absolute differences between its red component and the other two-color channels exceed a defined threshold, θ₂. This ensures that pixels with a strong red dominance characteristic of brown tones are effectively masked. The classification condition is expressed as:

Soil_Algae_i,j ← | R_i,j ​− ​G_i,j | ​≥ ​θ₂ AND | R_i,j ​− ​B_i,j | ​≥ ​θ₂

The threshold θ₂ can be adjusted based on the characteristics of the dataset, typically ranging between 1 and 10. For this dataset, θ₂ was set to 10, ensuring reliable identification and removal of brownish pixels associated with soil and algae. By applying these two hypotheses, all unwanted background pixels were successfully masked out of from the original color image, preserving only the RGB values corresponding to plant structures (Figs. S1A–B). This resulted in a refined image containing only green pixels, accurately isolating the plant. After applying the predefined thresholds to mask out unwanted pixels, an area filter was implemented to further reduce background noise and enhance segmentation accuracy. To facilitate precise extraction, the image was then inverted, converting plant regions to white and the background to black (Fig. S1C). This refined binary mask was subsequently utilized to extract plant structures from RGB images and accurately map plant regions in fluorescence and thermal imagery. The mask was multiplied with the target images, ensuring that background pixels multiplied by zero appeared as black pixels in the final output. Meanwhile, the extracted plant regions were retained and visually represented as green in RGB images and grey in fluorescence images. To accurately isolate individual plants, the connected component algorithm was applied, which identifies clusters of pixels that are connected based on the 8-neighbor rule (where connectivity is defined on a vertex basis). Each connected group of pixels is then labeled as a distinct plant. The bounding box for each plant was determined by extracting the minimum and maximum coordinates of the connected pixels, thereby defining the smallest enclosing rectangle that fully contains the plant area. Following segmentation, key morphological parameters such as projected surface area, maximal diameter, and compactness were computed based on the previously established definitions (Fig. S1D; [14]). In brief, the projected surface area of the plant was computed as the number of pixels in the connected object. A calibration factor is used to convert the area in pixels into metric units (square centimeters). To compute the diameter of the plant, an ellipse is fitted that completely holds the plant area inside; then, by considering the major axis of the ellipse, it computes the equivalent diameter of the plant. The compactness of the plant is calculated as the ratio of the projected area of the plant to the area of the convex hull covering the plant, the latter being the smallest convex polygon containing all given points. To calculate the different parameters originating from the fluorescence and thermal images, segmented plants are mapped onto the images and different measurements are then computed by applying standard statistics and mathematical formulas like mean grey values and standard deviation. Conversion to chlorophyll content is achieved by a species-specific factor. Thermal images have lower resolution; hence, a pixel matching conversion had to be performed to map masks based on RGB images onto thermal images. Combining images with subsequent data processing allows the recovery of a set of image-based calculated parameters. Table 2 lists all parameters discussed below.

Table 2.

Parameters and cameras of the MADI system.

Parameter	Image	Camera
Projected rosette area	RGB	Manta (Visible/NIR)
Rosette diameter	RGB	Manta (Visible/NIR)
Rosette compactness	RGB	Manta (Visible/NIR)
Chlorophyll content	Fluorescence	Manta (Visible/NIR)
Maximum quantum yield of PSII (Fv/Fm)	Fluorescence	Manta (Visible)
Operating efficiency of PSII (Fq’/Fm’)	Fluorescence	Manta (Visible)
Plant temperature	Thermal	FLIR (LWIR)

2.3. Plant materials

Arabidopsis thaliana ecotypes Columbia-0 (Col-0, N1092), Landsberg erecta-0 (Ler-0, NW20) Limburg-6 (Li-6, N1326), Wassilewskija-1 (Ws-1, N2223), Kaunas-0 (Kn-0, N1286), and C24 (N906) were purchased from the Nottingham Arabidopsis Stock Centre (NASC, Nottingham, UK). Seeds of weak ethylene insensitive 8-1 (wei8-1; [35]) acc synthase 2-1 (acs2-1)acs4-1acs5-2acs6-1acs7-1acs9-1amiRacs8acs11 (acs8x; [36]), jasmonate resistant 1-1 (jar1-1; [37]), abscisic acid insensitive 1-1 (abi1-1; Ler-0 background; [38]), gibberellic acid insensitive (gai-1; Ler-0 background; [39]), gibberellin 2-oxidase 7 (ga2ox7; SALK_055721; [40]), and uv response locus 8-6 (uvr8-6; SALK_033468; [41]) were also acquired from NASC. ethylene insensitive 2-1 (ein2-1; [42]) and deetiolated 2-1 (det2-1; [43]) were bought at the Arabidopsis Biological Resource Center (ABRC, Ohio, US). ethylene-free-2 (etf-2; quintuple loss-of-function in ACC oxidases; [44]) was donated by Li-Jia Qu. coronatine insensitive 1–2 (coi1-2; [45]) was a kind gift from Roberto Solano. repressor of ga-t2 (rga-t2) gai-t6 rga-like 1-1 (rgl1-1) rgl2-1 rgl3-1 (della5x; Ler-0 background; [46]) was donated by Liz Alvey. All mutant seeds were in Col-0 background, unless mentioned otherwise.

2.4. Experimental design of stress monitoring experiments

2.4.1. Case study 1: lettuce drought stress experiment

Lettuce seeds were planted and grown on peat pellets (Jiffy, Stange, Norway) in Araflats (Betatech, Gent, Belgium). Plants were fertilized once at the start of the experiment using 1/500 NPK (Wuxal, Aglukon, Düsseldorf, Germany) and watered twice every week with tap water by pouring in underlying trays until a level of approximately 1 ​cm was reached. Plants were grown in 16h photoperiod at 22 ​°C. At plant level, photosynthetic photon flux density (PPFD) was 300 ​μmol ​m⁻² ​s⁻¹. Imaging started 14 days after planting. From this day on watering was stopped to induce drought. For control plants, watering twice a week was maintained. Soil moisture was monitored daily using a SM150T soil moisture sensor (Delta-T Devices, Burwell, UK). Images were taken every 3 ​h. After 213.5 ​h (approximately 9 days), plants were re-watered. For calculation of growth parameters, measurements were done at each time-point. For estimating chlorophyll content, before dawn chlorophyll fluorescence values were used. Thermal images were taken around midday (9 ​h after dawn), a time point when stomata are usually open in response to light.

2.4.2. Case study 2: Arabidopsis salt stress experiment

Arabidopsis plants were sown on peat pellets in Araflats. Seeds were sown on peat pellets hydrated with tap water containing a 1/500 NPK (Wuxal, Aglukon) solution. Water was supplied once a week to approximately 1 ​cm height in the trays. Subsequently, plants were grown to rosette stage under a 16h photoperiod with 300 ​μmol ​m⁻² ​s⁻¹ PPFD. Imaging started 21 days after sowing (DAS; 10-leaf stage). Salt treatment started immediately after the first image acquisition and consisted of watering plants with tap water containing 0 (control), 150 (moderate stress) or 300 ​mM NaCl (severe stress), following the same watering regime. Leaf tissue was also sampled for destructive spectrophotometric chlorophyll analysis (see before).

2.4.3. Case study 3: ecotype-specific differences and effects of mutations in hormonal pathways

Different ecotypes (Col-0, Ler-0, Li-6, Ws-1, Kn-0, and C24) and mutants in hormonal pathways (wei8-1, acs8x, jar1-1, abi1-1, gai-1, ga2ox7, ein2-1, det2-1, etf-2, coi1-2, della5x) were sown on jiffy substrate and grown under a 16h photoperiod with 300 ​μmol ​m⁻² ​s⁻¹ PPFD. Water was supplied once a week to approximately 1 ​cm height in the trays. Imaging started at 21 DAS and was done daily until the end of the experiment. RGB images were taken around noon.

2.4.4. Case study 4: Arabidopsis UV-B treatment experiment

Col-0 and uvr8-6 were sown and grown as described before (Case study 2). Plants were grown under a 16h photoperiod with 300 ​μmol ​m⁻² ​s⁻¹ PPFD. The full spectrum D-light Papillon High Pressure sodium lamps and Master Color CDM-TP MW lamps have a spectral distribution ranging from 380 to 750 ​nm and therefore do not contain detectable UV-A nor UV-B light. Imaging started 22 DAS. UV-B treatment started 24 DAS on half of the plants. Plants were treated 4h daily starting 2h after dawn with 1.15 ​W.m-2 UVB coming from a Philips TL12 broadband lamp and filtered through two layers of cellulose acetate (Jürgen Rachow, Germany). Background white light was maintained at 300 ​μmol ​m⁻² ​s⁻¹ PPFD. Control plants were protected from UV-B radiation by a polycarbonate foil with cut-off at 380 ​nm (Jürgen Rachow, Germany).

2.5. Statistical analyses

All statistical analyses for the plant experiments were carried out in Graphpad Prism. After evaluation of (non-)violation of assumptions, one-way or two-way ANOVA followed by post-hoc Tukey HSD was carried out. When assumptions related to independency were violated, as was the case in most experiments where imaging was carried out and data was collected over time on the same plants, alternative tests were applied. In case of balanced datasets, statistical analyses were carried out using two-way repeated measures ANOVA using a Geisser-Greenhouse correction for violation of sphericity. Subsequently, post-hoc Holm-Sidak multiple comparisons tests were conducted. On the other hand, when datasets were unbalanced, two-way mixed models were carried out followed by post-hoc Holm-Sidak multiple comparisons tests. For Figs. S5A and C two-sample t-tests were carried out. A p-value of 0.05 was maintained for all statistical tests. Data are presented as means and error bars represent standard deviations (unless mentioned otherwise). Sample sizes are indicated in the figure legends and individual data points are depicted in the graphs.

3. Results

3.1. The MADI system

The MADI consists of a two-camera system mounted on a robotized platform displaceable in three dimensions (Fig. 1, Fig. 2, Fig. 3). One of the cameras acquires thermal images in the LWIR spectral region (Fig. 1D), while the other camera acquires images in the visible and NIR range (Fig. 2A). The camera for visible and NIR images is combined with specific filters in front of its lens, as well as an external box that serves to block out ambient light while imaging the specimens. The system, thus, allows sequential acquisition of RGB images, NIR images, fluorescence and thermal images in any desired order. Because of the overhead position and the focal depth of the cameras, the system is very suitable for seedlings and plants with a horizontal growth habit, such as those that are suitable for vertical farming practices, including leafy greens (e.g. lettuce, kale, spinach, etc.), microgreens and sprouts, or general rosette-forming plants including the model plant Arabidopsis. In the case of Arabidopsis, 38 individual rosettes can be monitored in one image, and the current growth room area allows 608 plants. For other plants, such as lettuce or spinach, maximum capacity depends on the developmental stage. For example, a maximum of four two-week-old lettuce plants can be imaged in a single frame, resulting in a maximum capacity of 64 plants in the growth room.

The general workflow for MADI-assisted plant analysis consists of experimental design, image acquisition, image preprocessing (generation of pixel ratio images), segmentation, mapping, analysis of parameters and archiving of results (Fig. 3). The complete control of the MADI system is done remotely using a custom software program, which allows sequential image acquisition and initial image preprocessing. After preprocessing, the images are ready for feature extraction. The VIS-IR camera captured images can be used for analysis of growth parameters (Table 2), including projected rosette area, maximal rosette diameter and compactness (Fig. S1). In addition, for fluorescence images and thermal images, mean pixel intensity is measured to calculate chlorophyll content, chlorophyll fluorescence and plant temperature. Combining images with subsequent data processing allows the recovery of a set of image-based calculated parameters. Table 2 lists all parameters discussed further.

3.2. Chlorophyll fluorescence to estimate chlorophyll content

Non-destructive measurement of chlorophyll levels is a vital tool to monitor plant health. Though analytical methods through the extraction of pigments and subsequent spectrophotometric measurements are highly accurate, they do not allow for kinetic studies over time to monitor plant growth. Here, we developed a fluorescence-based methodology that employs chlorophyll fluorescence to estimate chlorophyll content, similar to the proposition of Gitelson (1999) [30]. In the MADI system, the ratio of whole plant fluorescence at 730 ​nm vs 700 ​nm (F₇₃₀/F₇₀₀) upon excitation by blue light was employed. Arabidopsis leaves with varying chlorophyll levels, caused by developmental differences (e.g. young versus old leaves, green versus senescent), were imaged for chlorophyll fluorescence and associated with chlorophyll content obtained by extraction and spectrophotometric analysis (Fig. 4A, Fig. S2A). For chlorophyll contents ranging from 10 ​μg ​cm⁻² to 40 ​μg ​cm⁻² chlorophyll fluorescence values were within a linear range (Linear regression with β ​= ​0.01733, 95 ​% CI [0.01501, 0.01965], p ​< ​0.0001, R² ​= ​0.7393; Fig. 4A). However, at high chlorophyll content, the fluorescence derived parameter tends to saturate, thus leading to a lack of discrimination power at higher chlorophyll content (Fig. S2). To further explore the applicability of the fluorescence-based parameter to estimate chlorophyll content in vivo, especially in the higher range, leaves of a variety of species were harvested and photographed for chlorophyll fluorescence on the MADI system, and analyzed spectrophotometrically. While a big variation of chlorophyll content was observed through destructive analyses, the fluorescence-based parameter exhibited saturation around a value of 1.5. For example, young leaves of Peumus boldus, a tree native to central Chile, fell within the range of Arabidopsis values generated for the calibration line (Figs. S2B–C, Fig. 4A), while old leaves had much higher chlorophyll contents reaching over 90 ​μg ​cm⁻² and the associated fluorescence based parameter remained around 1.5 (Fig. S2B). Other subtropical species with high chlorophyll levels also showed a F₇₃₀/F₇₀₀ parameter of around 1.5, even though their chlorophyll levels ranged from 50 to 120 ​μg ​cm⁻². This means that the fluorescence-based parameter to estimate chlorophyll content is not universally applicable and caution should be taken when implementing, especially with high chlorophyll-accumulating species or tissues. Calibration curves for tomato and lettuce, two commercially relevant crops, were experimentally determined (Fig. 4B and C). For both species, the fluorescence parameter F₇₃₀/F₇₀₀ showed a full linear relationship in the entire range of chlorophyll content values derived from the tissue samples. To confirm that the F₇₃₀/F₇₀₀ can be used to evaluate stress-induced alterations in chlorophyll synthesis and breakdown, an experiment was conducted in which Arabidopsis plants were grown under standard light conditions or covered from light for five days to stimulate dark-induced senescence. Using this method, senescent leaves can be identified and distinguished from control leaves by the F₇₃₀/F₇₀₀ ratio, which produces clearly separated sets of values (Fig. 4D).

3.3. Case study 1: drought stress monitoring in lettuce

A kinetic analysis of growth parameters, temperature detection and chlorophyll content allow to accurately monitor plant responses to drought. In a first case study, the response of lettuce plants to drought stress by deprivation of irrigation was monitored. Lettuce is one of the most important leafy vegetable species cultivated and consumed worldwide [47,48]. Lettuce plants require high amounts of water for optimal growth and are quite sensitive to drought stress, in particular caused by their shallow root systems [49,50]. Two-week-old lettuce plants were subjected to drought treatment by withholding watering, while well-watered plants served as control (Fig. 5A). Plant growth, temperature and chlorophyll content were followed in time. The output of the automated image analysis yields values per plant, for example for the rosette area (Fig. 5B). From individual plant values, time point specific averages can be calculated and yield patterns for projected area (Fig. 5C), compactness (Fig. 5D), and diameter (Fig. 5E). In the first days of the experiment, a daily, steady and rhythmic increase in rosette area and diameter can be observed (Fig. 5). At around 168 ​h, the projected area and diameter of drought-treated plants dropped. At this point, the soil moisture reached 10 ​% and leaves started to wilt (Fig. 5A), which was reflected in the observed area and diameter difference with well-watered plants. Upon re-watering at 213.5 h, both area and diameter returned to the level they had before turgor loss, but plants did not reach the size of well-watered plants. In both well-watered and drought-treated plants, compactness showed a rhythmic, diurnal pattern, typically associated with leaf movement, in which leaves are more upright during the night (Fig. 5D). For the drought-treated plants, two trends were observed at the point of turgor loss. First, the rhythmic leaf movements visible as changes of the compactness of the plants were no longer noticeable. Secondly, compactness decreased in response to water deficit. Hence, turgor loss, monitored by a sudden drop in growth parameter values as well as loss of rhythmicity in compactness and overall increase may be useful as a cue to timely water plants avoiding drought-induced death.

Fig. 5.

Monitoring of parameters during drought stress of lettuce plants. (A) Soil moisture content (%). (B) Output for rosette area per individual plant as derived directly from the RGB images. (C) Average projected rosette area error bars indicate SD (n ​= ​8). (D) Average compactness. Error bars indicate SD (n ​= ​8). (E) Average diameter. Error bars indicate SD (n ​= ​8). (F) Average plant temperature measured with LWIR. (G) Average chlorophyll content measured with F₇₃₀/F₇₀₀. Statistical analyses were carried out with two-way repeated measures ANOVA with Geisser-Greenhouse correction (p ​< ​0.05) followed by Holm-Sidak multiple comparisons tests (p ​< ​0.05). For panel (F) the two-way RM ANOVA demonstrated that time (24.77 ​% of total variation), treatment (34.33 ​%), treatment∗time (27.92 ​%) and plant (11.14 ​%) all exhibited a p-value of p ​< ​0.05. For panel (G) the two-way RM ANOVA demonstrated that time (95.25 ​% of total variation), treatment∗time (2.148 ​%) and plant (0.5607 ​%) all exhibited a p-value of p ​< ​0.05, while treatment (0.01333 ​%) was not significant. Bars represent means and error bars represent standard deviations. Individual datapoints are also presented. Asterisks indicate statistically significant difference between drought-treated (D) and well-watered plants (WW) (∗, p ​< ​0.05; ∗∗, p ​< ​0.005; ∗∗∗, p ​< ​0.0005). Yellow marked region indicates period in which water was withheld (moment of rewatering [213.5 h]).

In the same experiment, leaf temperature of the plants was also monitored (Fig. 5F). When soil moisture dropped to 50 ​% (time point 126 h), a significant increase in plant temperature was observed (1 ​°C), and at a 30 ​% soil moisture (time point 126 h) all drought treated plants individually had higher temperatures than the well-watered ones. At the end of the treatment, when plants were experiencing severe drought, leaf temperature increased to 4 ​°C above that of control plants. These data are in accordance with lower transpiration in drought-stressed plants, likely through stomatal closure [15,16]. Therefore, a rise in plant temperature occurs much earlier than growth-related changes and can be used as an early warning sign for drought stress. Upon re-watering, leaf temperature rapidly decreased again, but remained higher than that of the control plants. In addition, we noticed that young leaves of drought-treated plants became relatively greener during the experiment. Chlorophyll content of younger leaves was followed by approximation via a central ring covering mainly young plant tissue but excluding mature leaf tissues. An increase in chlorophyll levels was detected over time in both treatments, probably related to increase photosynthetic activity of maturing leaves. However, it was also found that at low levels of soil moisture (less than 5 ​%), when turgor loss had set in the older leaves, chlorophyll levels rose slightly, but significantly, in young tissue (time point 213 h; Fig. 5G). These levels did not recover immediately after re-watering.

3.4. Case study 2: effects of different degrees of salinity on chlorophyll content

In contrast to most previously reported findings, where abiotic stress generally leads to a reduction of chlorophyll content, we identified a small but significant increase in chlorophyll after drought stress. Given the overlapping physiological responses in drought and salt stress, we set out to assess chlorophyll content fluctuations after salt stress using the F₇₃₀/F₇₀₀ parameter. Arabidopsis plants were grown for 2.5 weeks under control conditions and subsequently challenged with either mild (150 ​mM NaCl) or severe (300 ​mM NaCl) salt stress. Both WT Col-0 and the salt hypersensitive ein2-1 mutant [51] were monitored for growth alterations and chlorophyll fluctuations during 15 days. Salt stress resulted in a significant reduction of rosette growth (Fig. 6A, Fig. S3A). In addition, a reduction in compactness was observed at both salt concentrations (Fig. 6B). However, while growth retardation already occurred after 5 days of treatment, visible changes in plant compactness occurred later (12 days). Though ein2-1 mutant plants are hypersensitive to salt stress, no clear differences between WT and ein2-1 were observed in terms of growth reduction. Interestingly, ein2-1 mutants growing in the absence of stress were clearly more compacted than the WT.

Fig. 6.

Monitoring various degrees of salt stress uncovers chlorophyll hormesis. Arabidopsis Col-0 and ethylene-insensitive 2-1 (ein2-1) plants were grown until 17 days old and subsequently watered with different salt treatment solutions (0, 150, or 300 ​mM NaCl). (A) Projected rosette area (7 ​> ​n ​> ​12). (B) Compactness (7 ​> ​n ​> ​11). (C) Average chlorophyll content measured with F₇₃₀/F₇₀₀ for Col-0. (D) Average chlorophyll content measured with F₇₃₀/F₇₀₀ for ein2-1. Calculated chlorophyll values (μg.cm⁻²) based on fluorescence parameter can be found in Figs. S3B–C. (E) Concentration of total chlorophyll (Chla ​+ ​Chlb) as determined spectrophotometrically (n ​= ​4). (F) Concentration of total chlorophyll (Chla ​+ ​Chlb) as determined spectrophotometrically (n ​= ​4). For panels (A–D) statistical analyses were carried out with two-way mixed model (p ​< ​0.05) and post hoc Holm-Sidak multiple comparisons tests (p ​< ​0.05) showing significant differences within each timepoint. For panels (E–F) one-way ANOVA was carried out followed by post hoc Tukey HSD tests (p ​< ​0.05). Different letters represent groups that are significantly different. Bars represent means and error bars represent standard deviations. Individual datapoints are also presented.

Chlorophyll content assessment revealed a dual pattern in chlorophyll accumulation in WT plants, depending on the time. Initially, salt-stressed WT plants displayed higher values for F₇₃₀/F₇₀₀ (Fig. 6C) related to higher chlorophyll values after calibration (Fig. S3B). After a longer period of stress, F₇₃₀/F₇₀₀ significantly dropped compared to control conditions. In addition, this drop started earlier (12 days after treatment) and was greater for 300 ​mM NaCl, compared to 150 ​mM NaCl. Interestingly, such a hormetic response was absent in ein2-1 plants (Fig. 6D, Fig. S3C). Ethylene-insensitivity initially did not result in an increase in chlorophyll content, but led to a faster and stronger decrease in chlorophyll levels at both mild and strong salt stress. This is consistent with the previously reported hypersensitivity phenotype of ethylene-insensitive mutants [51]. To confirm the existence of hormesis for chlorophyll content, spectrophotometric determination of chlorophyll levels was conducted in parallel (Fig. 6E and F), revealing similar trends as those observed for F₇₃₀/F₇₀₀.

3.5. Case study 3: ecotype-specific differences and effects of mutations in hormonal pathways

We further investigated compactness of different mutants in hormonal pathways as well as various ecotypes (Fig. 7). Interestingly, considerate differences could be found in between hormone mutants. For example, at 21 DAS, several mutants deficient in growth hormones, such as the brassinosteroid synthesis mutant det2-1 and the auxin synthesis mutant wei8-1, were significantly more compact than WT (Fig. 7A and B). In contrast, jasmonate-insensitive mutants jar1-1 and coi1-2 were similar in size compared to Col-0, but grew less compacted with more slender leaves. For gibberellins, mutants with low levels of bioactive gibberellins (ga2ox7) or with constitutive repression of the signaling pathway (gai-1) were similar to their respective WT backgrounds in terms of compactness, while a mutant with constitutive gibberellin signaling (della5x) displayed decreased compactness. Lastly, we observed distinct morphological differences between plants deficient in ACC, the precursor of ethylene (acs8x), and plants lacking ethylene (etf-2) or blocked ethylene signal transduction (ein2-1). Plants deficient in ACC were much less compact compared to WT, whereas those deficient in ethylene were similar or more compact than WT. These results further support specific roles for the ethylene precursor ACC, as previously demonstrated [52,53]. Several ecotypes were also analyzed, with varying levels of compactness (Fig. 7A and B). To consider possible growth effects and developmental delays, plants were reanalyzed at the time of bolting (Fig. 7C), revealing similar trends.

Fig. 7.

Compactness differs between mutants in hormonal pathways and different genotypes. Arabidopsis WT plants and mutant alleles were grown on jiffy substrate and compactness was measured. The following genotypes in Col-0 background (orange) were included: Columbia-0 (Col-0) WT, weak ethylene insensitive 8-1 (wei8-1; auxin biosynthesis mutant), acc synthase 2-1(acs2-1)acs4-1acs5-2acs6-1acs7-1acs9-1amiRacs8acs11 (acs8x; ethylene biosynthesis mutant), ethylene-free-2 (etf-2; ethylene biosynthesis mutant), ethylene insensitive 2-1 (ein2-1; ethylene-insensitive signaling mutant), coronatine insensitive 1–2 (coi1-2; jasmonate-insensitive receptor mutant), jasmonate resistant 1-1 (jar1-1; jasmonate metabolism mutant), gibberellin 2-oxidase 7 (ga2ox7; gibberellin catabolism mutant), deetiolated 2-1 (det2-1; brassinosteroid biosynthesis mutant). The following genotypes in Ler-0 background (pink) were included: Landsberg erecta-0 (Ler-0) WT, abscisic acid insensitive 1-1 (abi1-1; abscisic acid insensitive signaling mutant), gibberellic acid insensitive (gai-1; gibberellic acid insensitive signaling mutant), repressor of ga-t2 (rga-t2) gai-t6 rga-like 1-1 (rgl1-1) rgl2-1 rgl3-1 (della5x; gibberellic acid constitutive signaling mutant). Lastly, the following ecotypes (green) were added in addition to Col-0 and Ler-0: Limburg-6 (Li-6), Wassilewskija-1 (Ws-1), Kaunas-0 (Kn-0), and C24. Images were taken around noon. (A) Average compactness of plants grown under long-day conditions until 21 days after stratification on jiffies. Plants were roughly the same developmental stage (8–10 leaves). Statistical analyses were carried out for each set separately. For comparisons in between genotypes in Col-0 background (orange), a one-way ANOVA (p ​< ​0.05) and post hoc Tukey HSD (p ​< ​0.05; lower case letters) were carried out. For comparisons between genotypes in Ler-0 background (pink), a Welch ANOVA (p ​< ​0.05) and post hoc Dunnett's T3 tests (p ​< ​0.05; lower case letters in italics) were carried out. For comparisons between different ecotypes (light blue, including Col-0 and Ler-0) a one-way ANOVA (p ​< ​0.05) and post hoc Tukey HSD (p ​< ​0.05; upper case letters) were carried out. 16<n ​< ​24, except for abi1-1 in which n ​= ​6. (B) Representative images of analyses presented in panel (A). Scale is 25 ​mm ​(C) Average compactness of plants grown under long-day conditions at time of start of bolting. Statistical analyses were carried out for each set separately. For each analysis, a one-way ANOVA (p ​< ​0.05) and post hoc Tukey HSD (p ​< ​0.05; lower case letters) were carried out. 12<n ​< ​24. Significant differences between genotypes in Col-0 background (orange) are indicated with lower case letters, those for genotypes in Ler-0 background (pink) are indicated with lower case italics letters, and those for the different ecotypes (green, Col-0, Ler-0) are indicated with upper case letters. Different letters represent groups that are significantly different. Bars represent means and error bars represent standard deviations. Individual datapoints are also presented.

3.6. Case study 4: UV-B effects on chlorophyll fluorescence parameters

Apart from the above-mentioned fluorescence-based chlorophyll estimation, chlorophyll-fluorescence-based parameters have been defined decades ago and, since then, frequently used in plant stress monitoring. We used the operating quantum efficiency F_q’/F_m’ and the maximum quantum efficiency of PSII F_v/F_m to follow the effects of UV-B (1.15 ​W ​m⁻²) on Arabidopsis WT and uvr8-6 mutant plants. The latter lack a UV-B specific photoreceptor responsible for photoprotective measures in plants [41,54]. UV-B treatment started on day three of imaging, 24 DAS.

Although a tendency is already visible the first day of exposure, significantly lower F_q’/F_m’ was observed after two days of treatment in the WT (Fig. 8A). In uvr8-6 mutants, a significant reduction was already observed after one day. The first F_v/F_m after exposure to UV-B was conducted the morning after the first treatment (day 25), and showed a clear reduction in both WT and uvr8-6 already after the first day of treatment (Fig. 8B). Notably, both F_q’/F_m’ and F_v/F_m were less affected in WT, which maintained relatively stable values over time, than in uvr8-6 mutants, the latter displaying progressively lower values for both parameters (Fig. 8A and B). Clear signs of UV-B damage were more visible in the RGB images of uvr8-6 plants (Fig. 8C), as well as in the chlorophyll fluorescence images. These results indicate that WT plants have a UVR8 dependent adaptation mechanism to photosynthetic stress caused by UV-B. This is in agreement with the role of UVR8 in driving UV-photoprotective measures, including the induction of flavonoid biosynthesis.

Fig. 8.

Chlorophyll fluorescence parameters of photosystem II change upon exposure to UV-B. Both 24-day-old Arabidopsis WT Col-0 and UV-B receptor mutant uv-b resistance locus 8-1 (uvr8-1) were treated 4 ​h daily, 6h after dawn. (A) Quantified average operating quantum efficiency of photosystem II (F_q’/F_m’) observed after the daily UV-B dose, 6 ​h after dawn. (B) Quantified average maximum quantum efficiency of photosystem II (F_v/F_m) detected at the end of the night. Statistical analyses were done by two-way repeated (RM) ANOVA with Geisser-Greenhouse correction (p ​< ​0.05) followed by Holm-Sidak multiple comparisons tests (p ​< ​0.05). For panel (A) the two-way RM ANOVA demonstrated that time (20.89 ​% of total variation), treatment (35.45 ​%), treatment∗time (24.48 ​%) and plant (13.59 ​%) all exhibited a p-value of p ​< ​0.05. For panel (B) the two-way RM ANOVA demonstrated that the factors time (5.218 ​% of total variation), treatment (28.88 ​%), treatment∗time (41.80 ​%) and plant (13.86 ​%) all exhibited a p-value of p ​< ​0.05. Different letters represent groups that are significantly different, ns: not significant. Bars represent means and error bars represent standard deviations. Individual datapoints are also presented (n ​= ​8). (C) Photographs showing Arabidopsis rosettes (25 DAS) in RGB (Red Green Blue mode), and the respective plants in working quantum efficiency (F_q’/F_m’) and maximum quantum efficiency mode (F_v/F_m). False color scales were applied to chlorophyll fluorescence images in ImageJ.

4. Discussion

4.1. The MADI system: comparison with other systems, advantages and limitations

In response to the critical challenges of climate change and the need to enhance agricultural productivity, high-throughput phenotyping has emerged as a vital tool for gathering comprehensive information on plant development and health across large populations. Real-time monitoring facilitated by these phenotyping platforms offers crucial support for crop management practices and aids in the identification of germplasms with desirable traits [5,[55], [56], [57]]. Employing various imaging technologies, from visible (RGB) imaging for growth monitoring to multispectral imaging for characterizing physiological responses, these platforms are integral components of modern agricultural research and automation efforts. The MADI system is an imaging platform that employs multispectral imaging to capture data on plant growth and physiology. While it shares some similarities with other phenotyping platforms, the MADI system also offers unique advantages and certain limitations, which will be discussed below.

Many current HTP platforms operate as plant-to-sensor systems, using conveyer belts to transport individual plants to dedicated imaging chambers. This ensures stable environmental conditions during imaging, minimizing variations in lighting, background, and other factors. Moreover, conveyor systems often accommodate larger plants. Notable examples of these systems include, but are not limited to, Phenoscope [10], PHENOARCH [11], HRPF [58], WIWAM Conveyor (PHENOVISION, [59]). In contrast, sensor-to-plant systems like the MADI, though less common, offer distinct benefits. They provide more stable environmental conditions throughout the experiment and limit manipulation of plants. However, they are sometimes constrained in maximum imaging capacity. Static systems on the other hand, like the DEPI platform [12], offer simultaneous imaging of large plant populations. However, these require multiple cameras and light sources, leading to higher costs and potential variations between sensors.

An important feature of the MADI system is its use of a filter wheel combined with the CCD camera, which enables imaging at different wavelengths, and can replace the use of multiple expensive sensors and allows for a certain degree of flexibility by introducing additional bandpass filters. The Enviratron at Iowa State University [60], for example, is a top-tier facility with numerous imaging capabilities facilitated by a combination of sensors, including a hyperspectral camera. These sensors are mounted on an unmanned rover that can visit different growth rooms with specific environmental conditions. Where Enviratron excels in its functionalities, it has a higher cost and a generally lower throughput compared to the MADI and similar systems. The design of the MADI therefore reflects a trade-off between throughput, cost and camera quality, incorporating high-quality CCD cameras with a good signal-to-noise ratio.

With the advent of machine learning and deep learning algorithms, data analysis of data gathered by hyperspectral cameras is becoming more feasible. Hyperspectral imaging collects data from a whole range of wavelengths [61]. Nevertheless, while hyperspectral imagers offer the potential for novel insights, their extensive wavelength range often includes redundant information. This redundancy increases computational demands, heighten the risk of overfitting, and complicate interpretation. Moreover, datasets with high dimensionality may impede effective modeling efforts, needing dimensionality reduction as a preprocessing step for hyperspectral image analysis [61]. Last but not least, despite certain advances, hyperspectral imaging provides spectral information of which the biological meaning has yet to be unraveled. For example, the recently developed Phenomenon platform combines RGB and thermal imaging with a micro spectrometer for the growth and health monitoring of in vitro plant cultures with high success [62]. Compared to the MADI system, the incorporation of high-resolution hyperspectral imaging has clear advantages in terms of possible data input. However, there is also a clear trade-off with the throughput. Acquisition of high resolution depth data scans is time-consuming, whereas the MADI system can complete a full imaging protocol on 38 Arabidopsis plants within minutes. It is important to consider the needs, possibilities and constraints upon designing novel or installing existing imaging platforms. Therefore, we chose to focus on imaging specific wavelengths, rather than the full spectrum. In the future, coupling a hyperspectral camera with advanced data processing tools could further enhance the MADI system's capabilities.

The MADI system builds upon on the strengths of existing systems. Most systems allow only limited imaging possibilities. Using two cameras and six filters, the MADI, on the other hand, allows the acquisition of multiple parameters, including growth-related measurements, chlorophyll fluorescence and thermal imaging. For instance, PHENOPSIS and Phenoscope platforms primarily utilize RGB images for high-throughput growth parameter assessment, while GROWSCREEN FLUORO and DEPI combine visible imaging with chlorophyll fluorescence to monitor both growth and photosynthesis continuously [12,63]. The MADI uses an updated methodology for the simultaneous measurement of growth and chlorophyll fluorescence. Various protocols exist for determining chlorophyll fluorescence parameters, with most platforms, including GROWSCREEN FLUORO [64], using PAM-fluorometry-based imaging from commercial manufacturers like PSI and Walz. The MADI system on the other hand utilizes the optimized DEPI protocol, first introduced by Cruz et al. [12], for chlorophyll fluorescence measurements. Furthermore, the MADI houses the CCD camera and LEDs inside an imaging box, allowing plants to be kept in temporary dark conditions for F_v/F_m measurements, even during daylight conditions. This feature is unique to the MADI system.

In addition to growth and photosynthesis-related parameters, the MADI system incorporates thermal imaging for leaf temperature measurements, and a fluorescence-based method for chlorophyll content assessments, making it one of the most integrated imaging systems available. While several systems such as Enviratron, NaPPI [65], and PHENOPSIS possess thermal cameras for temperature measurements, chlorophyll content determination often lacks, making the MADI unique in its kind. The system's filter wheel allows for future additional measurements, such as determining carotenoid or anthocyanin contents [66,67].

The MADI system is best suited for plants with a horizontal growth habitus, such as the model plant Arabidopsis, as well as important leafy vegetable crops like lettuce, spinach, and kale, among others. In this study, we demonstrated the utility of the MADI system in monitoring drought responses in lettuce (Fig. 5), serving as a case study for other leafy vegetables. Lettuce is a globally significant vegetable crop, commonly used in salads and as a source of nutrients, minerals and antioxidants [68]. In addition, lettuce is particularly sensitive to drought as it requires substantial amounts of water to sustain leaf growth [50]. In contrast, crops with a longitudinal growth habitus, such as most agronomically relevant monocots at adult stage, cannot be effectively imaged with the current system. For these crops, lateral imaging capabilities, such as those present in many plant-to-sensor platforms should be considered. Despite this limitation, the MADI system can still accommodate overhead imaging of young monocot seedlings, such as wheat or maize seedlings up to seven days of age. The limited focus depth of the MADI, however, restricts the imaging window to only a few days (Fig. S4). Nevertheless, the MADI could successfully be utilized for prescreening of large numbers of seedlings to isolate individuals with stress-resilient traits within genetic populations. To optimize the use of the MADI for monocot imaging, however, further optimization for accurate data extraction is essential.

4.2. Compactness as a bioassay for stress response

Identifying useful genetic variants, through large-scale screening of mutant populations or germplasms, requires the selection of meaningful, robust and useful markers. In the past, the use of well-designed bioassays has elucidated the genetic nature of signaling cascades and other molecular pathways. For example, most research on ethylene stems from studies employing the triple response phenotype, which is seen in etiolated seedlings treated with ethylene [[69], [70], [71]]. Its popularity originates from the fact that this bioassay is fast, easy, and cheap, making it ideal for screening of large populations of individuals [71]. In practice, it has led to the elucidation of nearly the entire canonical ethylene signaling pathway and revealed crosstalk with other hormonal pathways [72,73]. Similarly, a large number of molecular players involved in plant resistance to pathogen attack have been identified using bioassays monitoring lesion size [74]. Using the MADI system, we have demonstrated the utility of compactness as a bioassay to screen plants, either in various conditions or different genotypes, and their responses to stresses (Fig. 5, Fig. 7, Fig. 9). Stress often alters shoot growth architecture, typically leading to growth retardation or arrest as plants adapt to conserve energy and adjust metabolism (Figs. 5B and 6A; [75,76]). However, plants can also adjust their architecture, through leaf movements, petiole elongation or other morphological changes [77]. In a recent study, van Hooren et al. [78] revealed that Arabidopsis rosettes exposed to drought stress become less compact. Here, we showed that lettuce plantlets also display a decrease in compactness under drought conditions (Fig. 5C), demonstrating that this morphological response is not constrained to Arabidopsis. Moreover, the compactness parameter followed a circadian rhythm, as previously proven for Arabidopsis [78,79], which was lost under drought conditions, perhaps related to decreased turgor pressure or altered leaf movements upon drought stress (Fig. 5C). Together, these findings suggest that compactness serves as a valuable indicator for drought stress in lettuce, in addition to Arabidopsis, and may even be useful for monitoring other rosette-growing plants (Fig. 9). It can prove especially useful in situations where other imaging technologies, including thermal imaging or chlorophyll fluorescence images, which tend to respond faster, are not available. Solely by using RGB images, a simultaneous and cost-effective assessment of area and compactness can reveal the response of plants to drought or other stresses, without the need of complex imaging systems. In addition, distinct changes in compactness were found in various mutants in different hormonal pathways and ecotypes (Fig. 7). For example, the distinct response of ACC- versus ethylene-deficient plants supports the contention that ACC acts as a signal in addition to its role as a precursor of ethylene. These data highlight the usefulness of monitoring compactness for large-scale screening to identify interesting genotypes or germplasms in addition to the classical bioassays.

Fig. 9.

The MADI stress catalogue. Overview of the warning signals for plant stress, their associated imaging requirements, and the timepoint at which they can be observed. SM: soil moisture content, DAS: days after stratification. For more information on the different parameters see Table 2. Figure was made in BioRender.

Salt stress did not decrease compactness until the end of the experiment, in which loss of turgor and even wilting, probably caused the observed decrease in compactness (Fig. 6B). However, in previous studies, Arabidopsis was shown to be more compact growing under saline conditions [79,80]. In our case study, Arabidopsis plants displayed only a small non-significant increase in compactness after five days of treatment with 150 ​mM NaCl (Fig. 6B). In contrast, plants treated with 300 ​mM NaCl were comparable to the control treatment. The apparent disparity between our results and previous findings could be attributed to differences in dosage, plant age, and timing, which masked a stronger effect on compactness in our study. For example, both Awlia et al. [80] and Dhondt et al. [79] analyzed plants in an earlier developmental stage, and used 100 ​mM NaCl and 50 ​mM NaCl, respectively, as a treatment solution. In addition, the latter used an agar-based system. Further exploration of the effects of different stresses on compactness should consider dosage as well as age, and aim for a high temporal resolution. Interestingly, certain stress conditions are already known to induce morphological changes in shoots that affect stem and petiole elongation. For example, environmental factors such as shade [[81], [82], [83]], high temperatures [46], or submergence [84] can trigger elongation. Indeed, four-week-old plants exposed to ten days of low light (10 ​μmol ​m⁻² ​s⁻¹; Figs. S5A–B) or three days of submergence (Figs. S5C–D) revealed distinct elongation of leaves, which could be measured as a decrease in compactness. In such scenarios, assessing reductions in compactness can serve as a useful proxy for studying these plant responses or screening for genotypes with altered sensitivity to these conditions, in a high-throughput manner, saving time by omitting tedious petiole length measurements.

4.3. Non-destructive chlorophyll content imaging and other functionalities

While imaging of growth-related parameters, such as compactness and area, can prove useful, the advantage of the MADI system is its ability for combined assessment of growth and other physiological parameters which allows for a more complete view of plant health status. Moreover, depending on the goal and the conditions of an experiment (e.g. duration), changes in some parameters, rather than others, will be more evident. For example, under drought stress, lettuce plants responded very quickly in terms of changes in leaf temperature (Fig. 5F), whereas effects on growth took longer (Fig. 5B–E). For chlorophyll content determination, several methods already exist, including destructive analytical approaches [26]. Nevertheless, destructive methods do not allow for assessments over time on the same plants and are quite labor-intensive. Alternatively, methods based on chlorophyll fluorescence can also provide accurate determination of in planta chlorophyll content [30]. The latter utilizes the ratio F₇₃₅/F₇₀₀ to quantify chlorophyll content, which was modified for incorporation into the MADI system. Gitelson et al. demonstrated the linear proportionality of chlorophyll content to F₇₃₅/F₇₀₀ following excitation with 430 ​nm (blue) light. Our results with Arabidopsis revealed that the relationship between chlorophyll content and a modified F₇₃₀/F₇₀₀ parameter is not invariably linear; saturation occurs at higher chlorophyll levels within leaves (Fig. 4A, Fig. S2A), limiting the confident application of this non-destructive parameter within its linear range. Moreover, we observed species-specific variations in the relationship between chlorophyll content and fluorescence (Fig. 4, Fig. S2), suggesting the need for caution when utilizing this parameter across different species and calling for initial testing prior to application on other species. This lack of generalization over species has been reported for other chlorophyll content indices or methods to non-destructively estimate the chlorophyll content [[85], [86], [87]]. These discrepancies likely arise from differences in leaf optical properties, leaf thickness, chloroplast arrangement, which can be species-dependent but even age- or condition-dependent [86,[88], [89], [90]].

Furthermore, accumulating evidence indicates that chlorophyll content is dynamic and can change over the course of development [91]. Studies investigating the effects of abiotic stress on chlorophyll accumulation in plants show a negative correlation, with chlorophyll levels decreasing under stress conditions [[92], [93], [94]]. However, this response is not uniform; for instance, chlorophyll degradation upon stress often occurs faster in mature leaves, while younger leaves maintain chlorophyll levels longer to enhance survival [94]. Our study highlights chlorophyll accumulation in young leaves of lettuce rosettes exposed to drought (Fig. 5G). A meta-analysis incorporating data from over 20 different stresses, spanning 33 species and 177 dose-responses, revealed a stimulatory effect on chlorophyll levels under low to mild stress conditions, particularly within a limited time window [95], and an inhibitory effect at higher stress levels, a dose-response phenomenon called chlorophyll hormesis. In another study, when different barley varieties were subjected to salt stress at 150 ​mM NaCl some of these exhibited increased chlorophyll levels [96]. Similarly, in our study, 150 ​mM NaCl initially increased chlorophyll levels during stress (up to five days), while longer durations led to decreased chlorophyll levels compared to control conditions (Fig. 5C–E). This hormetic response of chlorophyll likely acts as a priming mechanism or conditioning strategy to cope with subsequent, more severe stress conditions, while at the same time ensuring survival under mild stress conditions. Together, these findings advocate for careful and time-dependent examination of chlorophyll accumulation, which is facilitated by high-throughput imaging systems like the MADI.

4.4. Conclusion and future perspectives

The MADI system represents a powerful tool for real-time monitoring of various plant growth parameters, including leaf temperature, photosynthetic efficiency, and chlorophyll content. Its high throughput and comprehensive approach offer valuable insights into plant development, physiology, and stress responses, revealing connections to the underlying molecular-genetic networks. This study not only demonstrates the effectiveness of the MADI platform for screening and growth monitoring, but also uncovers new aspects of plant phenotypic responses to stress, with compactness emerging as a key indicator. The high temporal resolution of growth alterations can further unravel the dynamics of both shoot architecture and morphology, as well as leaf movement, and can even identify novel genetic factors that contribute to it [78]. The potential to leverage genetic variation in traits like compactness could also pave the way for exploring hormonal and stress signaling pathways. Moreover, inclusion of other rosette morphology parameters, like eccentricity and roundness, could provide further insights into the timing of plant development and can shed light on the circadian control of growth [65,97].

The findings also offer guidance for future research on abiotic stress monitoring, outlining critical parameters and their stress-specific signatures (Fig. 9). Using the MADI system, the mechanisms of plant stress responses could thus be further elucidated, discovering the underlying physiological mechanisms and even uncovering novel biomarkers for specific abiotic stresses [98]. For example, drought stress assays and simultaneous dynamic monitoring of multiple growth and physiological parameters can further enhance our understanding of water relations and their connection to growth. Moreover, the MADI is particularly suited to study the relation with photosynthesis, through chlorophyll content and chlorophyll fluorescence measurements. The applications of the MADI in genetic research are also promising. The MADI allows for high-throughput investigation of genetic variation present in natural or mutagenized populations through, for example, GWAS studies or in forward genetics screens. Its capabilities are especially useful for multi-trait GWAS, in which various traits are incorporated in the study and analyzed with powerful computational models to enhance statistical power and identify pleiotropic loci [99,100]. In this, the MADI can be a robust and powerful tool for the identification of genes and alleles related to the regulation of shoot growth [101,102], physiological or developmental processes [78] and stress resilience [58,98,103], which are often resulting from complex genetic interactions. The MADI is, however, equally valuable for reverse genetics studies, particularly when large-scale screening of multi-gene families is warranted [65].

Although tested primarily on Arabidopsis and lettuce, the versatility of the MADI system suggests its applicability to other plant species, including other leafy vegetables or, for instance, Marchantia, a model plant for bryophytes, which also grows horizontally [104]. Overall, the MADI system underscores the value of a multifaceted imaging approach for understanding plant responses to stress, offering pathways to developing climate-resilient crops for the future.

Author contributions

AIVR and DVDS conceived the design of the system. AIVR and AVdB constructed the system. TD, FV and DVDS conceived the experimental work. TD, FV and RM performed the experimental work. TD and FV analyzed the data. RJQ aided in data analysis and processing. TD, FV and DVDS interpreted the data. TD, FV and DVDS wrote the first draft of the manuscript. All authors were involved in reviewing and editing of the final manuscript. DVDS coordinated the research and acquired funding.

Data availability

The data presented in this manuscript is freely available upon request.

Funding

DVDS gratefully acknowledges funding by the EU's Research and Innovation program Horizon 2020 (TimeScale), the Research Foundation Flanders (FWO, project G032717N), the Francqui Foundation for the Collen-Francqui Research Professorship (STI.DIV.2022.0014.01) awarded to her, and Ghent University.

Declaration of competing interest

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

Appendix A. Supplementary data

The following is the Supplementary data to this article: