A plants’ water and nutrients are primarily absorbed through roots, which in a natural setting is highly dependent on the 3-dimensional configuration of the root system, collectively known as root system architecture (RSA). RSA is difficult to study due to a variety of factors, accordingly, an arsenal of methods have been developed to address the challenges of both growing root systems for imaging, and the imaging methods themselves, although there is no ‘best’ method as each has its own spectrum of trade-offs. Here, we describe several methods for plant growth or imaging. Then, we introduce the adaptation and integration of three complementary methods, root mesocosms, photogrammetry, and electrical resistance tomography (ERT). Mesocosms can allow for unconstrained root growth, excavation and preservation of 3-dimensional RSA, and modularity that facilitates the use of a variety of sensors. The recovered root system can be digitally reconstructed through photogrammetry, which is an inexpensive method requiring only an appropriate studio space and a digital camera. Lastly, we demonstrate how 3-dimensional water availability can be measured using ERT inside of root mesocosms.
Introduction
Increasing agricultural productivity with fewer inputs is key to combat challenges posed by an increasing global population and climate change [1,2]. Agronomically important traits (e.g. yield, nutritional content) are strongly influenced by the root systems' ability to explore the soil, compete for water and nutrients, and adapt to their local environment. Indeed, several reports show the effectiveness of targeted breeding for root traits to enhance plant survival and efficiency [3–7]. Yet, despite their fundamental importance, little is known about root biology and RSA compared with above-ground structures. Difficulties in studying roots are inherently tied to their function; being imbedded in soil keeps them hidden from observation, and iterative branching growth rapidly increases their architectural complexity. Although studying root structure and behavior is not new [8], recent technological advancements have made describing the 3-dimensional (3D) nature of root structures feasible. Despite this, currently available methods are all imperfect, and in particular, there exists a wide gap between high-information content, controlled environment experiments, and limited measurements obtainable from the field. We outline the assets and deficits of several methodologies for both plant growth and imaging, then focus on the use of mesocosms to provide a compelling balance among physiological relevance, throughput, and completeness of the RSA. Finally, we present the pairing of mesocosms with unique applications of photogrammetry to reconstruct 3D RSA, and electrical resistance tomography (ERT) to measure 3D water content dynamics, collectively producing new insights into root system structure and function.
Growth methods for root imaging
Good root system measurements rely on the combination of plant growth, imaging, and computational analysis. Here, we focus on methods relating to plant growth and image capture; we direct readers to Atkinson et al. [9] for a review of computational methods. Each method allows the researcher to monitor specific traits which both restricts the experimental design and demands various amounts of labor, infrastructure, and know-how to conduct. To more easily parse the large number of methods, we reduced the number of techniques into coarse groups (Table 1). Each method was scored on several functional characteristics using TRUE/FALSE calls or subjective scores. The principal component analysis revealed four groups (Figure 1A) that are divided by imaging methods: 2D-optical imaging, photogrammetry, non-destructive field, and tomographic methods. Similarly, growth methods separated into three groups (Figure 1B) based primarily on where each method can be deployed: laboratory, greenhouse, or field.
A course grouping of growth and imaging methods using principal components derived from Table 1.
Binary characteristics called as TRUE (T)/FALSE (F) or scored 1–5, with higher scores being ‘better’ (a ‘labor’ score of 1 indicates an extremely labor-intensive method). Abbreviations: complete root system (CRT), computed tomography (CT), ground-penetrating radar (GPR), magnetic resistance imaging (MRI), positron emission tomography (PET).
State-of-the-art root phenotyping methods are a long way from achieving the ideal of comprehensive 3D root system structural and functional measurements over time in the field. Laboratory methods offer the least realistic conditions, but need little infrastructure. Methods such as 2D agarose plates (Figure 2A) have been used extensively, particularly in Arabidopsis, while 3D growth systems which leverage transparent media have also been used to great effect in crop species (Figure 2B) [10–12]. These approaches allow cheap, high-throughput, and non-destructive observations. Although several aspects of environmental complexity are caveated when interpreting results (e.g. ‘soil' physical, chemical, and biological properties, potential root phototropic issues [13]), they have been instrumental in studies of root growth, development and environmental response. To date, they have produced the only large-scale analyses of 3D and 4D RSA [12,14–16], including recent work indicating key characteristics can persist when comparing field-grown root crowns and gel-grown plants [15].
Visual representation of the growth methods.
Greenhouse methods provide more balance between the natural variance of a field and the consistency of a laboratory environment. Typically, this involves pot-grown plants (Figure 2C) which are excavated and photographed for minimal expense. While plants grown in potting mixes or reconstituted field soils provide a more realistic physiochemical environment than transparent media, disadvantages exist [17]. Increased temperature fluctuations, unrealistic water and nutrient availability, a reduced or absent soil microbiome, and physical constriction of roots may significantly affect RSA. Root mesocosms (Figure 2D), very large containers that allow essentially unconstrained root growth in a more realistic volume of soil, offer a compelling balance between realism and control, and are discussed in detail below.
Field methods trade information content for realism [18]. The most popular method is root crown excavation with a shovel (‘shovelomics' [19] (Figure 2E)). While intrinsically a small and superficial part of the overall RSA, the crown is the information-dense nexus. Root crown analysis has been effective across a diversity of research topics ranging from the effect of agricultural practices on root growth [20], drought-induced root traits [21], the identification of QTLs [22], and mutant analysis [23]. However, the desire for non-destructive measurements across the entire root system prompted the development of rhizotrons, large subterranean chambers where roots in physical contact with clear walls can be studied, requiring extensive long-term infrastructure [24–28]. Minirhizotrons, small diameter clear tubes that allow a camera to image roots touching the tube walls, were later developed and deployed en masse in the field. This innovation allows for multiple-timepoint sampling along a continuous depth profile [29], but considering a root system can explore dozens or hundreds of cubic feet of soil, information content is significantly limited [30]. Despite this, they have found important applications in many environments from crop fields to forests [31–33], including one of the world's most sophisticated root research facilities [34]. Soil cores offer a similar estimation of RSA as minirhizotrons, and although destructive, the recovered root and soil samples can be further analyzed [35–37]. Trench excavations and soil monoliths date back to at least the early 1900s [38–40] (Figure 2F) and offer the most complete data of any field method, but at an incredible infrastructure and labor cost. While modern incarnations have incorporated some basic image analysis [23,41], data acquired from trenches has been largely descriptive with coarse quantification of typically only a cross section of the actual 3D RSA.
Imaging methods for root systems
Example depictions for several of the imaging methods.
2D-optical imaging with a digital camera or flatbed scanner provides the most versatile and least expensive imaging method (Figure 3A–C). Although the 3rd dimension is sacrificed, entire root systems can be measured either in situ or excavated, intact or in pieces, depending on the approach. Many algorithms and software packages have been devised to extract complex features and traits from still images [39]. Programs such as DIRT [42,43], archiDART [44], EZ-Root-VIS [45], GiA Roots [46], WinRHIZO (Regent Instrument Inc., Ville de Québec, QC Canada), and RhizoVision [47] provide a means to measure of a variety of traits of interest; such as root lengths, diameter classes, volumes, and surface areas of individual roots or entire root systems of many species, for example, Zea mays [48].
Tomographic methods, which employ a series of 2-dimensional slices to generate a 3D volume, include X-ray tomography (XRT), magnetic resonance imaging (MRI), and positron emission tomography (PET) (Figure 3J–R). They provide non-destructive, high-resolution data of external and internal structures or the movement of small molecules through a root system. Adopted from medical and industrial imaging facilities, they typically require expensive equipment, highly trained users, and a dedicated space. While information content is high, including 3D and 4D growth and physiological dynamics, they are limited to relatively small-diameter pots, are moderately low throughput, and require extensive image processing and analysis. Nonetheless, advancements have been made recently in developing these tools for impactful plant science. Although differences in equipment and experimental parameters can vary, typically XRT has finer resolution (10s–100s μm) with lower penetration of the growth media, while MRI has lower resolution (100s–1000s μm) but accommodates larger pots [49]. Recently, advances have been made utilizing PET in conjunction with XRT or MRI to pair structure with function (Figure 3R). The movement of these radioisotopes can be detected in real-time at high-resolution (1–5 mm) and provide extraordinary power to the study water and nutrient uptake [50–52]. Another emerging technology is the use of neutron radiography (NR) to assess root structure and function. This has been used to not only reconstruct pot-grown root systems but also can provide high-resolution (10s–100s μm) 3D information of water distribution in the rhizosphere [53], as well as assess the individual water uptake capacity of specific roots in a single system [54]. Furthermore, NR has been paired with fluorescent dyes to image changes in rhizosphere oxygen content [55], and is hypothesized to be capable of assessing the dynamics of a variety of environmental factors.
The group ‘non-destructive field imaging’ includes two methods: minirhizotrons and ground-penetrating radar (GPR) (Figure 3D–F). Minirhizotrons capture 2D-optical images from cameras in buried tubes and are the only method, aside from dedicated rhizotron facilities, for observing time-course root traits in a field setting. GPR systems emit and detect radio waves that can identify underground roots [56]. While GPR is capable of surveying large plots, it is sharply limited by the minimum detectable root size (0.5+ cm) [57]. Hence GPR is an effective in forestry, but has limited applications for row crops without significant increases in resolution.
The methods presented here represent an array of creative solutions to the study of RSA. Advances in the applications of medical imaging technology to plant science have given us the ability to quantify detailed growth, morphology, and physiological processes at multiple scales. However, we thus far cannot capture the structure of mature root systems that reflect their in situ 3D configuration, nor can we easily collect concomitant dynamic environmental or physiological information in structures larger than a simple pot. In the following sections, we present the use of low-cost cameras and large root growth boxes (mesocosms) to generate 3D reconstructions of entire, unconstrained RSAs (Figure 3G–I), in combination with real-time sensor arrays.
Modern mesocosms for modern root studies
Mesocosms, medium-scale experimental systems that are meant to span the gap between the laboratory and field, offer aspects of both environmental control and realism. An old idea that spans many disciplines of biology, these systems vary widely even within the plant and ecological sciences [58–63]. Our concern was to preserve the 3D RSA of freely grown and mature crops, growing in physiologically relevant conditions, at a scale where multidimensional environmental gradients can effectively be measured and modeled (Figure 4). The mesocosms presented here are large growth containers (45 ft3 [1.3 m3]) which are scalable and allow for completely, or near-unrestricted root growth of full-size crop plants (Figure 4A). Within the container, several tiers of thin wire form a grid scaffold, and at harvest the root system is carefully excavated, preserving the 3D architecture. Multiple plant species and varieties have been grown and subsequently imaged via photogrammetry (discussed below), including switchgrass ecotypes VS16 and WBC3 (Figure 4B), from which 3D traits have been extracted (Figure 4C). The internal scaffold can also be used to isolate distinct spatial regions. Allowing for subsampling the system to match other phenotyping methods, such as isolating root crowns analogous to shovelomics samples. Figure 4E shows biomass collected in each cuboid at five z-levels providing a ground truth for computational approaches.
Root system mesocosms allow for both 3D reconstruction and 3D sensor data.
While multiple sensor probes can be deployed in any pot, the scale of traditional containers makes this additional data redundant. In a mesocosm, probes placed at multiple depths allow for measurements of biologically meaningful gradients in media matric potential, temperature, electrical resistance, CO2, et cetera (Figure 4F–I). Currently these gradients can be compared with the recovered root system to form post hoc prediction of root activity, with an aim to utilize one or more combination of sensor arrays to detect root activity in vivo.
Although the root mesocosm system is a powerful tool which does not require singularly expensive components, it does come at the expense of labor and space. A single mesocosm of the size described here costs a few hundred USD in lumber, nuts/bolts. Dedicated spaces and tools (pallet jacks) are important to store and handle the volume of media used in a mesocosms suite, and the physical labor required to un/load them is significant. Moreover, their weight requires reinforced concrete floors and substantial space to deploy, considering only a dozen mesocosms can cover a small (∼700 ft2) greenhouse. This is non-trivial, as establishing enough replicates to overcome the biological variance found in RSA may limit avenues of experimentation. As such, mesocosms are most useful as an in-between step from laboratory/greenhouse discovery to field testing, determining if juvenile traits continue onto maturity. If maintained, performance can be evaluated in the field. Furthermore, many other crops, such as fruits/vegetables or smaller cover crops, such as pennycress, could utilize significantly smaller units based on their RSA; potentially small enough to run large-sample size experiments in a greenhouse. Finally generating a point cloud that can be reconstructed though photogrammetry requires a dedicated studio which has multiple considerations in its size, location, and layout.
Leveraging photogrammetry for root system reconstruction
Photogrammetry, the use of 2D-photographs to create geometric reconstructions of 3D objects, is a powerful imaging methodology because it is decoupled from specific experimental environments, finding applications in the laboratory (gel imaging), greenhouse (mesocosms), and in the field (shovelomics). In a photogrammetry studio (Figure 5A) the sample is positioned to minimize shadows and allow for 360° imaging (Figure 5B). Images are collected from around the subject (Figure 5C) and the individual images are used to form a reconstructed 3D point cloud (Figure 5D), after which the background is removed via image processing to isolate the sample (Figure 5E). This point cloud is then cleaned semi-automatically (Figure 5F) and skeletonized into a network system (Figure 5G). This workflow from 2D projections to a 3D model is similar to other tomography-family methods. A key difference is the sensor moves freely relative to the sample during photogrammetry, rather than rotating around a fixed axis. As such, there are spatial restrictions (sample dimensions) to what can be imaged. During reconstruction individual voxels are mapped to a 3D space to produce a point cloud, then a surface mesh and/or a skeleton is derived to generate the defined root system object. At this point, many of the postprocessing algorithms initially designed for tomographic methods such as XRT [15,64] and MRI can be used as the 3D representations are interoperable. Additionally, this opens up other avenues of analysis, such as persistent homology to assess holistic growth patterns [65].
Steps in a photogrammetry workflow.
Projections from methods such as XRT allow the resolution of internal features, since X-rays penetrate the sample. Conversely, photogrammetry captures reflected light, so any internal structure, or any feature outside of line-of-sight cannot be simply reconstructed, as occurs in dense root crowns. Photogrammetry also relies on significant computational power and dedicated reconstruction software (e.g. Pix4D). Despite these requirements, photogrammetry offers the lowest financial barrier, a few hundred dollars for a standard digital camera and $4990 for a perpetual Pix4D license, the greatest sample-type flexibility, and when paired with mesocosms provide a novel view of RSA.
Electrical resistance tomography to map 4D water availability
ERT is a technique developed to observe subterranean structures and processes through electrical pulses emitted from electrodes [66]. Probes are laid in a line or encircling an object to obtain data on a specific transect, such as CO2 fluxes deep in underground pools [67], monitoring the rate of water infiltration into concrete [68], or to determine different geological layers across a survey site [69]. However, if electrodes are buried and arrayed in a grid, then 3D resistivity measurements can also be obtained [70] and have been used to observe the dynamics of sub-surface gas–liquid mixing [71], CO2 fluxes [72], and high-resolution variances in soil water content [73].
ERT has been used in multiple studies to assess the spatial distribution of electrical resistance influenced by water uptake correlated to root biomass [73–75]. Other interesting uses have been found, such as assessing the distribution of tree roots in woodlands [76], identifying preferred pathways of underground water flow, and providing information on relative levels of water uptake across a single tree's root system [77]. On the larger scale, ERT has been used to assess metrics such as water levels and soil types across field plots [73,78–80].
We used an ERT probe array (Figure 6A,B) to monitor electrical resistivity in a switchgrass mesocosm over the course of 42 days. Comparing 28, 38, and 42 days to 14 days after transplant allows us to visualize water-poor shells which are co-localized to known root masses (Figure 6C). While this has only allowed for estimations of gross root location, further refinement of probe layouts and densities could be used to develop real-time maps of root activity. Furthermore, if ERT sensor co-ordinates are paired with the internal co-ordinate system of the mesocosms the resultant 3D point clouds can be aligned with 4D resistivity fluxes. If expanded this could be used to effectively monitor fluxes under many experimental conditions; such as plant-to-plant below-ground competition, or relative water uptake of various roots of a single system. If scaled up it could be possible to holistically monitor the 3D water flux under entire field plots, pinpointing locations of specific changes in water dynamics in real-time, allowing for a targeted approach to water management and true measurements of below-ground physiological functions.
Example electrical resistance tomography data.
Conclusion
The study of root systems has progressed from marveling at the capacity of roots to penetrate nutrient-poor soils, seeking out nutrient-dense patches, to the generation of VR-ready root systems. While the heart of the problem in studying roots has not changed (they are underground, complex, and dynamic structures), the number and scope of solutions has grown substantially. Here, we summarized many tried-and-true methods such as trenches and agar plates to the most technologically advanced tomographic methods. While all these methods have clear strengths and weaknesses, none are capable of quantifying the 3D root structure of whole root systems grown unconstrained the way mesocosms paired with photogrammetry can. They can be scaled up to larger containers to study multi-plant root system interactions, and co-ordinate with above-ground phenotyping tools, while being relatively inexpensive and straightforward to construct. Furthermore, pairings with existing and developmental-stage sensors can provide an additional dimension of dynamic chemical flux. The necessity to improve our understanding of desired root traits cannot be understated with the threat of climate change on our doorstep. The methods outlined here provide a methodological backbone for researchers as we work towards developing lines which more efficiently harvest water and nutrients at both shallow and deep soil depths.
Competing Interests
The authors declare that there are no competing interests associated with the manuscript.
Author Contributions
T.D. conducted experiments. T.D., S.M., and C.N.T. wrote the manuscript. M.L. analyzed the matric potential/temperature data.
Acknowledgements
We would like to thank Dr. Bradly Carr for instruction and assistance with ERT, Michael Schoenewies for expert advice on photogrammetry, the authors of [81,82] for use of Figure 3F,N under the CC-BY creative commons license, and the authors of [83] for their personal permission to reproduce image 2F under the CC-BY-NC-ND license. Finally, Plant Physiology for the reuse of the image 3O from [84]. Images 2F and 3M were obtained via Shutterstock. This material is based upon work supported by the Department of Energy under Award number: DE-AR0000820 and the National Science Foundation under Award number: IOS-1638507.
Abbreviations
References
Author notes
These authors contributed equally to this work.