High resolution synchrotron imaging of wheat root hairs growing in soil and image based modelling of phosphate uptake


  • Samuel D. Keyes,

    1. Engineering Sciences, Faculty of Engineering and Environment, University of Southampton, Southampton, UK
    2. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Keith R. Daly,

    1. Engineering Sciences, Faculty of Engineering and Environment, University of Southampton, Southampton, UK
    2. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Neil J. Gostling,

    1. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    2. Centre for Biological Sciences, Faculty of Natural and Environmental Sciences, University of Southampton, Southampton, UK
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Davey L. Jones,

    1. School of Environment, Natural Resources & Geography, University of Bangor, Bangor, Gwynedd, UK
    Search for more papers by this author
  • Peter Talboys,

    1. School of Environment, Natural Resources & Geography, University of Bangor, Bangor, Gwynedd, UK
    Search for more papers by this author
  • Bernd R. Pinzer,

    1. Swiss Light Source, Paul Scherrer Institut (PSI), 5232 Villigen-PSI, Switzerland
    Search for more papers by this author
  • Richard Boardman,

    1. Engineering Sciences, Faculty of Engineering and Environment, University of Southampton, Southampton, UK
    Search for more papers by this author
  • Ian Sinclair,

    1. Engineering Sciences, Faculty of Engineering and Environment, University of Southampton, Southampton, UK
    2. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    Search for more papers by this author
  • Alan Marchant,

    1. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    2. Centre for Biological Sciences, Faculty of Natural and Environmental Sciences, University of Southampton, Southampton, UK
    Search for more papers by this author
  • Tiina Roose

    Corresponding author
    1. Crop Systems Engineering Group, Institute for Life Sciences, University of Southampton, Southampton, SO17 1BJ, UK
    • Engineering Sciences, Faculty of Engineering and Environment, University of Southampton, Southampton, UK
    Search for more papers by this author

Author for correspondence:

Tiina Roose

Tel: +44 2380592374

Email: t.roose@soton.ac.uk


  • Root hairs are known to be highly important for uptake of sparingly soluble nutrients, particularly in nutrient deficient soils. Development of increasingly sophisticated mathematical models has allowed uptake characteristics to be quantified. However, modelling has been constrained by a lack of methods for imaging live root hairs growing in real soils.
  • We developed a plant growth protocol and used Synchrotron Radiation X-ray Tomographic Microscopy (SRXTM) to uncover the three-dimensional (3D) interactions of root hairs in real soil. We developed a model of phosphate uptake by root hairs based directly on the geometry of hairs and associated soil pores as revealed by imaging.
  • Previous modelling studies found that root hairs dominate phosphate uptake. By contrast, our study suggests that hairs and roots contribute equally. We show that uptake by hairs is more localized than by roots and strongly dependent on root hair and aggregate orientation.
  • The ability to image hair–soil interactions enables a step change in modelling approaches, allowing a more realistic treatment of processes at the scale of individual root hairs in soil pores.


Crop productivity is significantly constrained by poor bioavailability of sparingly-soluble nutrients, especially phosphate (P) (Barber, 1984; Tinker & Nye, 2000; Vaccari, 2009). The efficiency of P acquisition, in which root hairs play a critical role, is important in addressing global food-security issues that arise from increasing world population and climate change. Advances in genetics provide the capability to breed plants with improved root-hair phenotypes, with manipulation of length and spatial patterning providing potential routes to enhanced P uptake (Gahoonia & Nielsen, 2004; Datta et al., 2011). Discovery of optimal traits for P uptake is particularly important because rock P resources are predicted to be exhausted within 50–100 yr, meaning alternatives to traditional fertilization are imminently required. Despite the importance of root hairs to P uptake, the absence of published studies imaging undisturbed hairs growing in situ within soil has hindered the validation of mathematical models describing their function in the soil.

The proliferation of root hairs along major root axes has been implicated in a range of processes including enhanced nutrient and water uptake, root-microbial signal exchange, allelochemical release and plant anchoring (Bates & Lynch, 1996; Ehrhardt et al., 1996; Datta et al., 2011). Despite their ubiquity throughout the plant kingdom, the role of root hairs in plant–soil processes remains poorly understood due to the difficulties inherent to studying their behaviour under field conditions. This poor understanding has limited the targeted selection of root hair traits, particularly for enhanced nutrient acquisition in the field (Gahoonia & Nielsen, 1997, 1998, 2004; Gahoonia et al., 1999; Datta et al., 2011).

Previous morphological, image-based and genetic studies of root hairs have been conducted either in artificial hydroponic/agarose gel systems, by destructive washing of roots from soil (Ma et al., 2001), or cryo-scanning using an electron microscope (Watt et al., 2005). Clearly, hydroponic/agarose growth conditions are radically different from those existing in soil (e.g. in terms of mechanical resistance) and it is therefore unclear how data from these studies translates into real soil environments. Though experiments with mutants of various species in real soils have linked root hair density and length to P uptake, the hair-scale mechanisms of uptake are indistinguishable using these approaches (Bates & Lynch, 2001). Washing of soil-grown roots is limited as a technique due to the small scale and fragile nature of root hairs (c. 3–6 μm, see review in Leitner et al. (2010)), making it difficult to estimate the degree of loss or damage during the washing process. Resultingly, observation of root hairs growing in situ in real soil requires the use of non-invasive methods such as X-ray computed tomography (CT), using either a laboratory microfocus scanning system (mCT), or synchrotron radiation source (SRXTM). μCT scanners produce X-rays for imaging by bombarding a metal target with thermionically emitted electrons, producing a polychromatic, incoherent beam of photons in a cone or fan arrangement. Though satisfactory for many imaging applications, μCT systems have their limitations, particularly in terms of available X-ray flux. Whilst a variety of commercial microfocus scanners now operate at sufficient resolution to image structures at the root hair scale (c. 1 μm), scan durations at such resolutions, with good contrast-to-noise ratios, are inevitably in the order of hours. During such lengthy scans, movement due to plant growth and/or soil changes (e.g. local water movement) almost inevitably compromise image quality, making it problematic to use microfocus scanners for root hair imaging. By contrast, SRCT uses X-rays generated by electromagnetic deflection of an extremely high-energy (relativistic) electron beam. A photon beam produced by a contemporary synchrotron source is typically highly monochromatic, coherent and has a far greater photon flux than that of a microfocus source. These characteristics allow acquisition of radiographs at high resolutions, with adsorption and/or phase contrast, using greatly reduced exposure times. The consequent reduction in scan times down to the order of c. 5 min allows small, unstable structures to be imaged without gross intra-scan movement. Furthermore, phase contrast effects can be exploited to enhance edges between materials of similar adsorption contrast. For these reasons, SRCT was used in this work to image the root hairs of wheat plants grown in soil at a resolution of c. 1 μm.

Modelling of P uptake by root hairs has been hampered by a lack of reliable data about the geometric configuration of hairs with respect to soil particles, with the result that most models contain unsubstantiated assumptions, reducing their reliability and accuracy. Previous models (Barber, 1984; Tinker & Nye, 2000; Ma et al., 2001; Leitner et al., 2010) all present cases in which soil is treated as a continuum around the root hairs, sharing the underlying assumption that root hairs are much larger than soil aggregates. As can be seen from Synchrotron Radiation X-ray Tomographic Microscopy (SRXTM) imaging of root hairs and soil particles in this study (Figs 1 and 2), this assumption is false. Zygalakis et al. (2011) and Leitner et al. (2010, for hydroponic growth) do not make this assumption, instead using multi-scale homogenization theory to derive model equations for the situation where root-hair spacing is comparable to soil aggregate size. These models take the influence of soil pore-space on P diffusion into account, including not only soil aggregate influences, but also geometric hindrance arising from the root hairs (Leitner et al., 2010; Zygalakis et al., 2011). Whilst multi-scale homogenization provides a very useful tool for understanding the impact of small periodic structures on the large-scale behaviour it fails to provide sufficient information regarding factors influencing root-hair P uptake at the soil-pore scale. Optimization of models at the whole-plant scale will require validation of pore-scale uptake behaviour, including whether hairs or roots dominate P uptake, or whether root and hair contributions are comparable (Leitner et al., 2010). This study presents a complete work-stream and methodology for imaging living root hairs using SRXTM and translating resulting image data for assessment of plant P uptake using image based modelling.

Figure 1.

Region selection and classification for rhizosphere simulation from synchrotron data. (a) A segment is defined, with centreline aligned with the centre axis of the root, (b) All other voxels are removed, leaving only the defined segment, (c) root hairs, soil, fluid and root surface regions are individually defined using different discrete grey-levels, (d) a volume mesh is generated, with root hairs, soil, root surface and water defined separately.

Figure 2.

Digitally rendered 3D volume with the soil phases partially cut away to reveal c. 3 mm section of a seminal root of Triticum aestivum, including lateral roots and root hairs.

Materials and Methods

Plant growth microcosm

Winter wheat seeds Triticum aestivum (cv Santiago; KWS, Thriplow Cambridge, UK) were pre-germinated on wetted tissue paper for 48 h. Seven 1 ml syringe barrels were clustered, inserted into a larger tube of 3 cm diameter, and filled with soil, including a c. 5 mm layer covering the barrel openings. The soil used was a Dunnington Heath series Chromi-Abruptic Luvisol from the University of Nottingham field site at Sutton Bonington (52°49′52″, −1°15′13″). A single pre-germinated wheat seed was then planted in each assembly, such that individual roots could grow into the syringe barrels (Supporting Information Methods S1 and Fig. S1). Plants were grown for 10 d in a controlled growth environment (Fitotron SGR, Weiss–Gallenkamp, Loughborough, UK). Day conditions (16 h) were 23°C and 60% humidity and night conditions (8 h) were 18°C and 55% humidity, with both temperature and humidity ramped between states over a 30-min period. Plants were initially top-watered with 5 ml of water, following which 3 ml was added every 48 h to maintain water at close to field capacity.


Following growth, CT imaging was carried out at the TOMCAT Beamline on the X02DA port of the Swiss Light Source at the Paul Scherrer Institute, Villigen, Switzerland. Plant and soil material protruding above the barrel openings was excised using a scalpel, and individual syringe barrels were separated from the cluster immediately before scanning at 20 keV, with 1501 equiangular projections acquired through 180°. The projections were post-processed to generate light and dark corrected sinograms, which were in turn converted to a stack of two-dimensional (2D), 8 bit tiff files for each scan. Features were extracted from data using a combination of automated tools for soil classification, and manual segmentation for root hairs using a graphical tablet and AVIZO Fire 7 software. The resulting segmented regions were used to generate volume finite element (FE) meshes for COMSOL Multiphysics using ScanIP.


In COMSOL the uptake of P by root surfaces and desorption from the particle surfaces was modelled, based on previous work (Schnepf et al., 2011). Though the idealized geometry used previously (Schnepf et al., 2011) was simpler than the situation in this study, their mechanistic approach represents the closest model in the literature to that required for image based modelling, that is, it considers P uptake and spatially distributed binding reactions on soil particle surfaces, rather than soil bulk binding reactions.

The sub-region used to generate the FE model for simulations was defined according to the following procedure. The constraints of available computational power required that the model consist of < 9 × 106 elements. Choosing meshing parameters which preserved soil surface morphology, the maximum allowable sub-region volume was found to be c. 1 × 10−18 m3. For reliable comparison with future data, a segment geometry was selected as a repeatable geometry, with centerline aligned to that of the root. A number of constraints determined the positioning of the segment for the simulation. The image region was required to: be unaffected by visible blurring artefacts; exclude lateral roots; exclude edge effects at the soil/tube boundary. In the regions satisfying these criteria, soil structure was relatively homogenous as a result of preparation method. Since hairs were also evenly distributed over the root surface, the specific segment position (determined randomly) is qualitatively representative of the soil and hair characteristics of the entire dataset. Following definition, a multi-region, FE volume mesh was generated from the segment RAW data using the +FE-Free algorithm in ScanIP (Simpleware Ltd, Exeter, UK) for input to COMSOL (Version 4.2a, COMSOL Inc., Burlington, MA, USA).

The following equations and boundary conditions were applied to the FE model.

Diffusion of P in soil solution:

display math(Eqn 1)

where Cl is the concentration of P in the soil pore water measured in mol m−3, and D is the P diffusion constant in water in m2 s−1.

Mass conservation on soil surface is given by the following first-order reaction:

display math(Eqn 2)

where Ca is the amount of P bound per unit surface area of soil aggregate in mol m−2, ka is the rate of sorption in m s−1, and kd is the rate of desorption in s−1.

The root hair (and root) P uptake boundary condition is described by Michaelis Menten kinetics:

display math(Eqn 3)

where Cmin is the soil solution P concentration when the P uptake by plant stops, Fm is the maximum rate of P uptake in mol m2 s−1, Km is the concentration when uptake is half of maximum possible in mol m−3, and math formula is a unit vector normal to the root surface, pointing into the root domain, that is, all surfaces in three dimensions have two unit normals pointing in opposing directions; the only difference being the sign. We assume that the root and root hair P uptake properties are the same all over the surface. This might not necessarily be true since P transporters can potentially be heterogeneously distributed, but at this time we are unable to quantify such characteristics, and thus do not take heterogeneity into account. However, due to the spatially explicit nature of our model it would be very easy to include if and when experimental data becomes available.

The boundary condition at the soil particle surface boundary reflects the binding reactions at that location. Hence it is intrinsically coupled to Eqn 2, that is,

display math(Eqn 4)

where math formula is a unit vector normal to the soil aggregate surfaces.

We take a zero flux boundary condition on the external surfaces of the domain of integration:

display math(Eqn 5)

Initially the soil and associated P surface density and concentration are assumed to be in equilibrium, that is, math formula, where Ca0 and Cl0 are the initial concentrations on the soil surface and in the fluid, respectively. Though the quantity Cl0 must be derived experimentally, only the volume concentration, Csor, in the soil can be directly measured. We thus convert it into a surface concentration using Ca0 = CsorρV/S, where V is the soil volume in m3, S is the soil surface area in m2 and ρ is the bulk soil density in kg m−3.

Soil chemical characterization

Adsorption and desorption constants (ka, kd) for the simulations were parameterized using standard soil tests (Murphy & Riley, 1986; Giesler & Lundstrom, 1993) which determined the fractions of bound and soluble P (Supporting Information Methods S1, Fig. S2). The methods used to define these parameters are fully described in the Supporting Information. The bulk total binding rates were measured and the bound mass was distributed evenly to all imaged soil aggregate surfaces as the initial condition for the numerical simulation. Parameters for root and hair P uptake were obtained from the literature (Barber, 1984; Tinker & Nye, 2000). A list of all parameter values used for simulation are given in Table S1.

Post-processing of simulation results

The pore-scale simulation results were post-processed in Paraview and VGStudio MAX, and surface integration was conducted in COMSOL Multiphysics.


Representative examples of unprocessed image data are shown in Figs 1 and 2 (and also see Supporting Information Methods S1, Fig. S3, and Movie S1). These reveal the intricate and varied nature of root hair and soil particle interactions. In some instances, root hairs traverse large macro-pores, while others infiltrate intra-aggregate micro-cracks. The crucial question arising is what the geometric interplay between root hairs and soil aggregates implies about root and root hair P uptake from the soil; especially given that significant amounts of P are bound to the soil aggregate surfaces (Tinker & Nye, 2000). We used three-dimensional (3D), image-derived root, root hair and soil aggregate geometries to compute P uptake by the root and root hairs in a qualitatively typical sub-region representing 550 μm of root length over an angular range of c. 75° (Figs 1, 2) whilst taking into account dynamic binding on the soil particle surfaces and diffusion in the soil pore space. The FE simulation results (Fig. 3) were then analysed to determine P transport characteristics in the soil pore space.

Figure 3.

(a) Estimated surface concentrations of phosphate (Ca) on the soil particle surfaces after 10 h of uptake by root and root hairs of wheat (Triticum aestivum), (b) streamlines showing phosphate transport paths from soil surfaces to root hairs (magenta) and root (blue).

Initial quantities of bound P on the soil were parameterized using the agronomic indexing system adopted by the UK government to regulate use of P fertilizers (Defra, 2010). Index P3 implies that the total amount of P in the soil is 39 mg P l−1, which should be sufficient to produce crops, rendering additional fertilization inadvisable. Indices P2 and P1 reflect soils with lower total levels of P, 23 and 12 mg P l−1, respectively. Simulations were run for all three different agronomic soil P indices, assuming that the soil P binding properties were the same in each case. Initial concentration was also assumed to be homogenous over all soil surfaces. For all three P index scenarios, cumulative total P uptake from both root and all root hairs was comparable, with uptake by both decreasing to very low values after 5 h (Fig. 4). However, the uptake contribution by the root was c. 15% higher than for root hairs in all cases.

Figure 4.

Consolidated simulation results (over 10 h). Left axis indicates cumulative phosphate (P) uptake, plotted separately for root surface and root hairs of wheat (Triticum aestivum), with Ca0 values corresponding to phosphate indexes P1, P2 and P3. Right axis shows ratios of polyline distance between groups of streamline start points (on soil surfaces), and groups of corresponding end points (on root and root hair surfaces), where streamlines follow maximum Cl concentration gradients (shown in Fig. 3b). Plotted points are mean values for all sets, calculated separately for root and root hairs (standard error in means indicated by error bars).

Though the predominant paradigm in the literature (Barber, 1984; Tinker & Nye, 2000; Ma et al., 2001; Leitner et al., 2010) is that root hairs dominate P uptake. This result confirms that P uptake by root hairs and the root are intimately coupled. This calculation is with respect to the single root scale, but the behaviour at the root system scale might differ, since not all root surfaces are likely to be covered by hairs. Hence on the root system scale, the importance of the hairs might be reduced, especially in the later stages of development when the root system is more mature and thus more extensive. The area of the root surface in the simulation was 1.96 × 10−7 m2 and the total area of root hairs was 1.66 × 10−7 m2, such that the difference in P uptake is attributable to the difference in the total surface area rather than to the soil volume into which the hairs penetrate. This is a significant finding as it contradicts previous studies which concluded that root hairs dominate P uptake (Barber, 1984; Tinker & Nye, 2000; Ma et al., 2001; Leitner et al., 2010).

Plotting diffusion streamlines following paths of maximum concentration gradient from the soil particles to root and root hair surfaces allowed the flux of P to be visualized (Fig. 3b). Mapping depletion zones on soil particle surfaces to the corresponding root and root hair uptake surfaces allowed the degree of depletion gradient localization to be quantified. Sets of seed points were defined on the soil surface with inter-seed separation length δ0 = 0.25 μm, and the paths of maximum concentration gradient from seed point to uptake surface were computed. This allowed P regions on the soil surfaces to be mapped to specific sites where the released P was absorbed by the plant. Neglecting any jumps between different soil aggregates the average distance between streamline termination points on the soil surface is δL. The ratio of distances between streamlines on the soil aggregate and the root or root hair surfaces can now be estimated using R = δL/ δ0. Similarly the average length of the streamlines can be easily obtained by summing the arc lengths along the streamline trajectories. The mean ratios for the streamline sets, together with the error in the mean for both root and root hair surfaces is shown in Fig. 4. The sensitivity of the calculations to δ0 was measured by halving δ0 and repeating the calculations for a subset of the sampled points. This produced no noticeable difference in the results for average streamline length or distance ratio. The data show that root hairs draw P from a particle surface region whose area is roughly comparable to between two and three times that of the uptake area on the root hair surface. By contrast, the root extracts P from a much larger soil surface area of c. 5–10 times that of the root surface. This occurs because the root surface creates large global diffusion gradients in soil pores by comparison to the smaller gradients very close to soil particle surfaces created by root hairs. Root hairs have a high length to diameter ratio and are generally spatially closer to the surfaces of soil particles, such that associated P fluxes are more localized and influenced to a greater extent by hair diameter and curvature. A comparison of mean diffusion path lengths for the root surface and root hairs confirmed that dependent on time, soil surface to root paths were c. 5–9% longer than those from soil surfaces to the root hairs. Additional insight arises from the dynamic nature of the error bounds on the mean measure of the soil particle to root and root hair area ratio. Variation is initially greater (though it decreases with time), since during the early stage diffusion interacts dynamically with the binding of P to the soil particle surfaces, that is, the uptake-driven diffusion profile in the soil pore domain has begun to influence the soil surfaces, but the surface binding reactions have not yet reached equilibrium. It is also evident that after the soil particle surface-binding reactions reach equilibrium, there is a level of adjustment between the surface concentrations (via particle to particle interaction) at c. 2–3 h, after which the entire soil pore and soil surface domain reaches equilibrium (Fig. 4).


This study has demonstrated that in situ imaging of living roots in soil is possible using SRXTM, and that consequent resolution of individual root hairs with respect to soil aggregates can be achieved. Through use of suitable protocols and imaging methodology, the workflow from living plant to modelling results has been established for nutrient uptake for the first time. Assumptions must still be made regarding initial distribution of P, but a step change in modelling has been brought about by removing the assumption of idealized geometry.

Clearly, all scientific methodology is party to limitations, and this is true for our work presented here. In particular, to achieve the spatial resolution necessary for imaging root hairs, it was necessary to develop a growth assay that was small (c. 5 mm in diameter). Due to the physical constraints involved in imaging with X-rays, this requirement for small sample sizes is unlikely to change radically in the near future. Additionally, even a very powerful computer enabled us to simulate only a small section of the entire volume imaged. We envisage that this particular constraint will disappear in the very near future due to rapid improvement in software and hardware, such that within 1–2 yr, the possibility of running image based simulations on the entire imaged domain can be envisaged. Finally, there are specific ways in which validation of the model could be achieved. One particularly promising approach for validation of our P modelling results is the use of X-ray fluorescence tomography, which should in theory allow mapping of the distribution of P in the soil. Though promising, the technique is still in its infancy, but our growth assay is highly suitable for investigation as soon as the technique becomes fully accessible.

Thus, the image data acquired, combined with modelling in this study, have been utilized to substantially improve upon existing P uptake models, showing that P uptake by root hairs is a rapid process, occurring on the same timescale as P uptake by the root. Quantification of the zones from which root hairs extract P from the soil shows that the majority is extracted from soil surfaces immediately adjacent to root hairs, with very little acquired from particle surfaces oriented away from them. These results have major implications for our understanding of how root hairs should be modelled at the plant and crop scale, and indicate that previous modelling studies should be re-visited using an imaging-based, multi-scale homogenization approach. Understanding of P uptake grounded in hair-scale mechanisms can now be used to optimize plant models at larger scales, where resolution of single hairs is impossible. This research also opens up possibilities for assessing how different soil surface binding reactions and geometries might influence different rhizosphere processes. Furthermore, the methodology developed in this study will now enable investigations of rhizosphere processes, involving many replicates of different plant and soil treatments. It also raises the possibility of imaging live mycorrhizal hyphae in the soil together with host roots. Mycorrhizal hyphae are of similar size to root hairs (c. 10 μm in diameter), but the structure is much more complex (Smith & Read, 2002). We believe our assay could be extended and modified to address questions of mycorrhizal–plant interaction, thus enabling the validation of plant mycorrhizal models (Schnepf et al., 2005, 2008a,b, 2011; Schnepf & Roose, 2006; Thonar et al., 2011). Another area where our technology could be applied is the assessment of mucilage and root exudates in the rhizosphere; thought to be highly important in P uptake (Hinsinger et al., 2005). Whilst X-ray CT is not necessarily suitable for imaging mucilage and exudates due to their near-identical density to that of water, other methods such as neutron tomography are well suited to such investigations. The plant growth assay outlined here could be applied to this technique, which like X-ray CT has been hampered by the small sample sizes necessary to achieve the spatial resolution needed for rhizosphere studies and model development (Esser et al., 2010).


T.R. is funded by The Royal Society University Research Fellowship, K.R.D., N.J.G. and A.M. are funded by BBSRC, D.L.J. and P.T. are funded by DEFRA-Link, S.D.K. and I.S. are funded by EPSRC.