Our results demonstrate that gaps appeared after transpiration rate started to decrease. The decrease in transpiration was explained by the low hydraulic conductivity of soil at matric potentials lower than −5 kPa, as also reported by Carbon (1973). The reduced soil hydraulic conductivity limited the water flow into roots and caused the initial shrinkage of roots and the gap formation. Gaps limited even more the water flow into root, roots shrank further, and, as in a chain reaction, the water flow into roots became more and more limited until the plants finally wilted. Gap formation seems therefore a consequence and not the cause of limiting water flow from soil to roots as suggested by Faiz and Weatherley (1982): “Root contraction could hardly initiate a rise in the interfacial resistance since an initial increase in resistance is necessary to bring about a fall in water content of the root tissue and hence a reduction in root diameter”. However, it might be that gaps smaller than the spatial resolution were already present before transpiration decreased. Transpiration rate started to decrease at a soil matric potential between −5 kPa and −10 kPa. At such a matric potential, gaps larger than 0.03 mm (gap diameter) would be drained and could limit water flow. Although a higher resolution is required to exclude this hypothesis, there are two arguments suggesting that air-filled gaps larger than 0.03 were not present: 1) if such gaps were present, the gray values in the voxels at the root-soil interface would have been decreased and some of these voxels would have been classified as gaps; 2) independent experiments with neutron radiography (Carminati et al. 2010; Moradi et al. 2011) showed that during a drying period the water content in the rhizosphere of lupins was higher than in the bulk soil. This result was more pronounced for lateral roots and was explained by mucilage exuded by roots. The observed high water content in the rhizosphere and the hypothesized role of the mucilage suggest that roots were in “good” hydraulic contact with the soil before transpiration rate started to decrease. Similarly, root hairs may close the gap between soil and root, either by directly taking up water or by creating a capillary bridge for water to flow across the gap. White and Kirkegaard (2010) found a correlation between gap size and density of root hairs, suggesting a potential role of root hairs in the adaptation of plants to the presence of gaps and macropores. However, we cannot exclude that some partial contact and narrow gaps occurred earlier at some locations. All that we can conclude is that the large and continuous gaps between the tap root and soil appeared after the decrease in transpiration and were not the cause of the decreased transpiration.
The tomograms show a vertical gradient in gap formation around the tap root, with larger gaps at the basal part and smaller gaps towards the apical part. Gaps initiated near the root collar, that lost contact to the soil before the apical part of the tap root – compare gap size at top and bottom of Lupin I and IV. As soil matric potential decreased (Lupin II and IV), the gap size along the tap root was proportional to the root radius. The gap was equal to the root shrinkage and relative root shrinkage was quite uniform along depth. The uniform shrinkage of the tap root could be explained by a quite uniform xylem potential, a probable consequence of high xylem conductivity, and by a uniform ratio of stele and cortex cross sectional area, with the latter being more susceptible to shrinkage.
Much smaller gaps were observed around lateral roots, which shrank by a maximum of 9 %. This result is at the limit of, and probably below, our spatial resolution, and needs to be verified with higher spatial resolution. However, the possibility that laterals remain in a closer contact with the soil compared to the tap root has important implications for soil-plant water relations and deserves discussion. Why did lateral roots shrink less than the tap root?
One possibility is the different elasticity of laterals and tap root. Laterals may have a smaller cortex in proportion to the total cross section. Since cortex shrinks more than the stele in lupin, this would result in a smaller relative shrinkage of lateral roots.
A second possibility is that the xylem water potential in lateral roots was higher than in the tap root. This could be caused by a limited xylem conductivity of the laterals. If the xylem vessels of the laterals were not yet mature, their axial hydraulic resistance could be significant. The axial resistance of laterals could be further increased by unfavorable connections between laterals and tap root, as reported by Byrne et al. (1977) in soybean.
An additional possibility is related to water flow across the root-soil interface. Figure 8 illustrates a conceptual model of root shrinkage based on the ratio between root and soil conductivity. When soil is wet (t1, Fig. 7) and its conductivity is high, the largest gradients in water potential occur across the root radial pathway. As soil water content decreases (t2), steep gradients in water potential arise between bulk soil and root surface because of the non-linear decrease of soil conductivity and the radial geometry of the water flow. The average water potential across the radial pathway becomes more negative. If cortex cells have no osmotic adjustment, their turgor pressure will start to decrease and shrinkage of cortex cells begins (t3). For a relative shrinkage of 1 %, a tap root with radius of 1,370 μm will form a gap of 14 μm, while a lateral with radius of 190 μm will form a gap of 2 μm. The air entry values for such gaps are approximately −20 kPa and −150 kPa, respectively. Hence the initial gap around tap root will be air filled at higher matric potentials and the hydraulic connection between soil and root will be lost earlier. Due to the reducing conductivity as gaps become large, the process will be self enhancing – i.e. once a gap is initiated, water uptake can occur only through vapour phase, the water potential in the radial pathway will become more negative and cortex cells will undergo further shrinkage. Given the same relative shrinkage, the initial gap will be larger around tap roots than laterals and it will be drained at a less negative water potential. Thus, the self-enhancing process will start earlier for tap roots and tap roots will shrink more, also in relative terms.
A final hypothesis why laterals may shrink less than the tap root, is the higher concentration of mucilage around laterals and more distal parts of roots. In recent papers, Carminati et al. (2010) and Moradi et al. (2011) observed increasing water contents towards roots. This result, contradicting the common paradigm of decreasing water content towards roots, was explained by the high water holding capacity of mucilage. This effect was more pronounced around laterals and in the distal parts of roots. As shown in a modelling exercise by Carminati et al. (2011), mucilage attenuates the gradients in water potential in the rhizosphere and consequently will reduce the loss of turgidity of root cells. Higher concentration of mucilage around laterals could therefore explain the different shrinkage of tap root and laterals and could help lateral roots to remain in contact with the soil.
To date, researchers have regarded gaps between soil and roots as a negative process for plant-soil water relations. However, air-filled gaps will partly isolate the roots from soil. For plants exposed to dry soils this may imply lower water loss from roots to soil, which in this case will occur by vapour diffusion, as suggested by North and Nobel (1997). Gap formation around the tap root, in particular in the top soil, and persistence of the contacts at the laterals may be a good strategy to enable plants to isolate parts of the roots that are in the dry soils, while younger roots continue to grow in wetter regions. Similarly, gaps and isolation of the most proximal root parts may not necessarily induce a reduction in root water uptake, as this can be compensated by more apical parts, as suggested by Garrigues et al. (2006) and Zwieniecki et al. (2003).