Outside-Xylem Vulnerability, Not Xylem Embolism, Controls Leaf Hydraulic Decline during Dehydration1[CC-BY]

Changes in leaf outside-xylem properties drive leaf and whole-plant hydraulic decline with dehydration, protecting plants from catastrophic embolism in xylem conduits. Leaf hydraulic supply is crucial to maintaining open stomata for CO2 capture and plant growth. During drought-induced dehydration, the leaf hydraulic conductance (Kleaf) declines, which contributes to stomatal closure and, eventually, to leaf death. Previous studies have tended to attribute the decline of Kleaf to embolism in the leaf vein xylem. We visualized at high resolution and quantified experimentally the hydraulic vulnerability of xylem and outside-xylem pathways and modeled their respective influences on plant water transport. Evidence from all approaches indicated that the decline of Kleaf during dehydration arose first and foremost due to the vulnerability of outside-xylem tissues. In vivo x-ray microcomputed tomography of dehydrating leaves of four diverse angiosperm species showed that, at the turgor loss point, only small fractions of leaf vein xylem conduits were embolized, and substantial xylem embolism arose only under severe dehydration. Experiments on an expanded set of eight angiosperm species showed that outside-xylem hydraulic vulnerability explained 75% to 100% of Kleaf decline across the range of dehydration from mild water stress to beyond turgor loss point. Spatially explicit modeling of leaf water transport pointed to a role for reduced membrane conductivity consistent with published data for cells and tissues. Plant-scale modeling suggested that outside-xylem hydraulic vulnerability can protect the xylem from tensions that would induce embolism and disruption of water transport under mild to moderate soil and atmospheric droughts. These findings pinpoint outside-xylem tissues as a central locus for the control of leaf and plant water transport during progressive drought.

Leaves account for the bulk of photosynthetic productivity and transpirational water use, and given the increasing incidence and severity of droughts in many regions (Vicente-Serrano et al., 2014;Diffenbaugh et al., 2015) if not globally (Sheffield et al., 2012), the mechanisms underlying the drought responses of leaves are ever more critical to understand. Reduction of photosynthesis and growth under mild dehydration and subsequent death under prolonged drought are primarily related to failure of the water-transport system (Tyree and Zimmermann, 2002;Sack et al., 2016a). Water moves under negative pressure through plant xylem, via the cohesiontension mechanism (Dixon and Joly, 1895), and a certain level of tension can cause air to aspirate through a xylem conduit, causing spontaneous vaporization, a process known as cavitation. The resulting embolization of the xylem conduits has been widely believed to be the main cause of hydraulic decline during drought (Milburn, 1966;Tyree and Zimmermann, 2002), which results in declines in gas-exchange rates Brodribb and Holbrook, 2003;Hernandez-Santana et al., 2016) and can ultimately precipitate plant mortality (Choat et al., 2012). While embolism is a major cause of the failure of stem hydraulic function, its role in leaves has not yet been clarified. Understanding the role of embolism on leaf hydraulic function is equally if not more important than in stems, as leaves represent a hydraulic bottleneck (Sack and Holbrook, 2006) that can determine plant hydraulic responses and the resulting declines in stomatal conductance and photosynthesis during drought (Brodribb and Holbrook, 2003;Sack and Holbrook, 2006). Leaves are highly vulnerable to dehydration, with leaf hydraulic conductance (K leaf ) often declining rapidly between full turgor and turgor loss point and even more strongly during extreme dehydration (Brodribb and Holbrook, 2006;Johnson et al., 2009b;Scoffoni et al., 2012;Sack et al., 2016b). This response could arise in one or more of several tissues, as water moves first through the vein xylem, then exits the xylem through bundle sheath cells and flows through the mesophyll before evaporating into the intercellular air space, and then diffusing through stomata out of the leaf ( Fig. 1; Tyree and Yianoulis, 1980;Boyer, 1985;Rockwell et al., 2014). Thus, the decline of K leaf with dehydration may be driven not just by reduced vein xylem hydraulic conductance (K x ) but also by reduced outside-xylem hydraulic conductance (K ox ), which includes pathways through the vascular parenchyma, the bundle sheath, and the rest of the mesophyll tissues. Both K x and K ox determine K leaf : Indeed, recent studies have suggested that cell shrinkage with dehydration, and/or deactivation of membrane aquaporins outside the xylem, could strongly reduce K leaf (Kim and Steudle, 2007;Shatil-Cohen et al., 2011;Pantin et al., 2013;Scoffoni et al., 2014;Moshelion et al., 2015;Sade et al., 2015). Yet, the vulnerability of K x and K ox , and their influences on K leaf decline with dehydration, have not been clearly disentangled. Although recent evidence has suggested that the leaf xylem is resistant to embolism under moderate levels of dehydration (Scoffoni and Sack, 2015; Bouche Figure 1. K leaf characterizes the water-transport capacity of the whole leaf and is influenced by water movement through the leaf xylem (K x ; A) and through the mesophyll, or outside-xylem pathways (K ox ; B), which includes vascular parenchyma, bundle sheath, and mesophyll cell pathways for liquid and/or vapor phase transport and diffusion through air spaces (red dots) through stomata. As the leaf dehydrates, observed declines in K leaf have typically been attributed primarily to reduction of K x due to the formation of embolism in xylem conduits, although recent studies suggested a possible role for changes in outside-xylem pathway properties via reduced membrane permeability and cell shrinkage. Symbols are as follows: xylem (X), bundle sheath cell (BS), spongy mesophyll cell (SM), palisade mesophyll cell (PM), upper epidermal cell (UE), lower epidermal cell (LE), and stomata (S).
et al., Brodribb et al., 2016a), whole-leaf hydraulic decline with dehydration has been most often primarily attributed to embolism, based on indirect evidence (Milburn and Johnson, 1966;Crombie et al., 1985;Kikuta et al., 1997;Salleo, 2000, 2003;Salleo et al., 2000Salleo et al., , 2001Nardini et al., 2001Nardini et al., , 2008Bucci et al., 2003;Lo Gullo et al., 2003;Stiller et al., 2003;Trifilò et al., 2003a;Brodribb and Holbrook, 2005;Woodruff et al., 2007;Johnson et al., 2009aJohnson et al., , 2012Blackman et al., 2010Blackman et al., , 2014. For instance, the earliest report of xylem embolism was for leaf petioles, based on acoustic emissions thought to be caused by cavitation events (Milburn, 1966), and subsequent studies reported that the number of acoustic emissions a leaf generates correlated with leaf hydraulic decline (Tyree and Sperry, 1989;Johnson et al., 2009a). However, it is now recognized that acoustic emissions from drying leaves may arise from processes other than xylem conduit embolism, such as fractures in the tissues or embolism within fibers or mesophyll cell walls (Sandford and Grace, 1985;Ritman and Milburn, 1988;Cochard et al., 2013). In severely dehydrated excised leaves, embolisms can be observed in the leaf vein xylem using scanning electron microscopy of cryogenized sections, dye methods, or direct light transmission, and several studies reported that K leaf decline corresponded to the accumulation of leaf vein embolism (Cochard et al., 2000;Trifilo et al., 2003b;Woodruff et al., 2007;Johnson et al., 2009a;Brodribb et al., 2016b) and suggested this to be the main driver of K leaf decline. However, there has been a lack of information of the number of embolized xylem conduits within given vein orders across the range of leaf water stress and their influence on K leaf (Wylie, 1947;McKown et al., 2010;Sack and Scoffoni, 2013) relative to the potentially strong role of vulnerability of the outside-xylem pathways. Recent work has proposed that outside-xylem hydraulic decline may play a role in K leaf decline (Sade et al., 2014;Scoffoni et al., 2014;Hernandez-Santana et al., 2016;Trifiló et al., 2016). A study that partitioned the vulnerability of K leaf into that of K x and K ox (Trifiló et al., 2016) found that both contributed, depending on species, but measurements were made under low irradiance, which would minimize the response of K ox before the turgor loss point (Guyot et al., 2012;Sack et al., 2016b). A strong test of the relative roles of K x and K ox depends on their determination for illuminated leaves coupled with direct observations of the formation of emboli in the xylem.
To test the relative roles of xylem embolism and changes in outside-xylem properties in determining the decline in K leaf during dehydration, we combined three approaches. We first investigated whether embolism occurred in leaf veins as leaves dehydrated to turgor loss and beyond using x-ray microcomputed tomography (microCT). We then quantified the vulnerability of K x and K ox to dehydration, which allowed us to partition their influence on the vulnerability of K leaf at any point during dehydration under high irradiance. We investigated the anatomical determinants of the decline in outside-xylem pathways using a spatially explicit model of leaf water transport. Finally, we tested the implications of our findings, using a model of the whole-plant hydraulic system to estimate the influence of the measured declines of K x , K ox , and K leaf , on wholeplant hydraulic conductance under different drought scenarios.

RESULTS
The main determinant of K leaf decline in dehydrating leaves was hydraulic vulnerability of the outside-xylem pathways, rather than xylem embolism, for eight angiosperm species from eight families ( Table I). The strong declines of K leaf during progressive dehydration above and below the turgor loss point did not reflect patterns of xylem embolism observed in vivo (Figs. 2 and 3). MicroCT imaging of dehydrating leaves of four species revealed few gas-filled conduits even at the turgor loss point when K leaf had already declined by over 60% (Figs. 2 and 3), where, on average, only 5% to 8.5% of midrib conduits were embolized across species in the midrib and none in the minor veins (Table II). Substantial levels of embolism (a maximum of 44% across species) were observed in the midrib only under extreme dehydration beyond the turgor loss point (Table II), but emboli were nonexistent or rare in the minor veins of these species at those extreme water potentials (Scoffoni et al., 2017). Hydraulic measurements of K x vulnerability across the four species used for microCT imaging and an additional four ecologically diverse species (Table I) corroborated the microCT evidence of low K x vulnerability on average across species compared with K leaf . Thus, the water potential inducing 50% loss of hydraulic conductance for the leaf Figure 2. Low vulnerability of the leaf xylem to embolism before the turgor loss point as revealed by in vivo imaging of leaves of four diverse angiosperm species subjected to progressive dehydration (i.e. increasingly negative leaf water potential [C leaf ]) using microCT. Scans show leaf midribs at mild dehydration, the turgor loss point, and extreme dehydration (an illustrative image for each range is shown from left to right), showing very few embolized midrib conduits above the turgor loss point. No emboli were observed in higher order veins above the turgor loss point, and few were observed even in extremely dehydrated leaves (data not shown). Note that C. diversifolia contains embolized protoxylem conduits, which are hydraulically nonfunctional, even for well-hydrated leaves, and these protoxylem conduits are included in the calculations of embolized conduits. Bars = 250 mm. xylem (P 50,Kx ; obtained from the K x vulnerability curves shown in Supplemental Fig. S1) was, on average, 1.6 MPa more negative than that for the whole leaf (P 50,Kleaf ; Fig. 3), representing a much lower sensitivity to water stress of K x than of either K leaf or K ox (P values of 0.015 and 0.007, respectively, by paired Student's t test for each species; values for P 50,Kx , P 50,Kox , and P 50,Kleaf are shown in Table I). By contrast, the water potential inducing 50% loss of hydraulic conductance for the outside-xylem pathways (P 50,Kox ) was, on average, 0.1 MPa less negative than P 50,Kleaf , representing only a slightly greater sensitivity. Although the vulnerability of K x to dehydration was much smaller than that of K ox for all species, their relative sensitivities varied: the P 50,Kx ranged from only 0.08 to 0.8 MPa more negative than P 50,Kox in two soft-leaved shrub species (Lantana camara and Salvia canariensis) to 2.9 to 3.2 MPa more negative in sclerophyllous species of the California chaparral (Comarostaphylis diversifolia and Quercus agrifolia). Partitioning the contributions of xylem and outside-xylem pathways to the decline of K leaf (see "Materials and Methods") showed that, across species, the decline in K ox explained 86% to 100% of the decline in K leaf at the turgor loss point (96% on average across species), 95% to 100% of that at P 50,Kleaf (98% on average), and 75% to 100% of that at water potentials inducing 88% loss of leaf hydraulic conductance (P 88,Kleaf ; 93% on average; Table III). Furthermore, while across species both P 50,Kx and P 50,Kox correlated positively with P 50,Kleaf (r 2 = 0.57 and 0.99, respectively), when testing models predicting P 50,Kleaf from P 50,Kox and/ or P 50,Kx , the model with P 50,Kox alone was selected by maximum likelihood as the better predictor (Supplemental Table S1), explaining 81% of P 50,Kleaf variation across species according to independent effects analysis.
Our model simulations of the plant hydraulicstomatal system showed that, on average, across species ( Fig. 4), and for three of four species individually (Supplemental Fig. S2; Supplemental Table S2), decline of K ox would be the main determinant of the decline of not only K leaf but of whole-plant hydraulic conductance under a wide range of scenarios of atmospheric drought (i.e. VPD) or soil drought (i.e. increasingly negative soil water potential [C soil ]). Indeed, the trajectory of the percentage loss of conductivity of the whole-plant hydraulic system to either type of drought showed strong overlap with that of K ox , while the bottleneck imposed by low K ox shielded the leaf and stem xylem hydraulic conductances from tensions that would result in significant declines in these components under increasing VPD or increasingly negative C soil . Roots also have water flowing through living tissues of the outsidexylem component, and root hydraulic conductance (K root ) shows steep hydraulic vulnerability (Brodribb and Hill, 2000;Hacke et al., 2000;North et al., 2004), but K root too is shielded from decline under increasing VPD by the bottleneck imposed by declining leaf K ox . Notably, like the other compartments, K root strongly declines under more negative C soil . However, because Figure 3. The vulnerability of K leaf (green lines) to dehydration is determined mainly by the vulnerability of the outside-xylem pathways (K ox ; dashed black lines) and not that of the xylem (K x ; light-gray lines) across the four species for which microCT was performed (left) and an additional expanded set of four diverse species (right). The maximum likelihood function is plotted for each vulnerability curve (see "Materials and Methods"). The turgor loss point for each species is represented by the vertical dotted black line.
C soil is less negative than leaf water potential during transpiration, K root does not decline as strongly as leaf K ox on average across species. Even for L. camara, which has a relatively vulnerable xylem, under increasing VPD, the decline of K ox is steep and protects the other compartments of the plant from high tensions as for the other species, although under soil drought, steep declines in hydraulic conductances would occur in all organs (Supplemental Fig. S2).
Across species, the vulnerability of the hydraulic pathways correlated with the drought tolerance of the mesophyll cells. Thus, bulk leaf turgor loss point (C TLP ) correlated with P 50,Kox and P 50,Kx (r 2 = 0.69 and 0.91, respectively; P # 0.01).
We applied model simulations to refine hypotheses for the source of the decline of K ox in dehydrating leaves. We parameterized the MOFLO model for water transport outside the xylem  with shifts in leaf anatomy and physiology that can be observed directly or that were determined experimentally or hypothesized in the literature to occur during dehydration, including leaf and internal tissue shrinkage, cell wall shrinkage, reduction in cell connectivity, and decreases in membrane permeability (Sancho-Knapik et al., 2011;Shatil-Cohen et al., 2011;Pou et al., 2013;Scoffoni et al., 2014;Sade et al., 2015), and with or without assuming an apoplastic barrier at the bundle sheath, as has been reported for some species (Lersten, 1997;Taneda et al., 2016). Across all four species, a reduction of membrane permeability in the context of an apoplastic barrier was the only factor that could directly account for the decline of K ox values during dehydration. Model simulations showed that an 80% reduction in membrane permeability in the context of an apoplastic barrier resulted in a 58% to 86% decline of K ox values. However, without an apoplastic barrier, the decrease of K ox due to membrane permeability reduction would not be important enough to overcome the opposing effect of tissue shrinkage. Notably, leaf and tissue shrinkage as measured from microCT images ( Fig. 5) would, by itself, actually increase K ox by 4% to 55% across species, by shortening flow pathways outside the xylem (Fig. 6). Furthermore, an 80% reduction in cell connectivity had little impact, and in most cases (especially under the "no apoplastic barrier" scenario), its decrease was not sufficient to overcome the increase in K ox induced by cell shrinkage (Fig. 6). Notably, an 80% reduction in cell wall thickness yielded reductions in K ox regardless of simulating an apoplastic barrier or not, with 11% to 72% declines in K ox at the turgor loss point across species and scenarios.

Vulnerable Outside-Xylem Pathways Protect the Xylem from Embolism throughout the Plant
Our results from both microCT imaging and hydraulics experiments suggest that the primary determinant of K leaf decline in leaves from mild to extreme dehydration originated in vulnerability of the outside-xylem pathways and not hydraulic failure of the xylem. Across species, the decline in K ox caused more than 85% of the decline in K leaf by the turgor loss point and more than 75% by P 88,Kleaf . These results are consistent with the body of literature linking changes in aquaporin expression to leaf hydration status and bundle sheath and mesophyll cell turgor (see below; Johansson et al., 1998;Steudle, 2007, 2009;Miyazawa et al., 2008;Shatil-Cohen et al., 2011;Shatil-Cohen and Moshelion, 2012;Pou et al., 2013;Prado and Maurel, 2013;Laur and Hacke, 2014;Scoffoni et al., 2014;Sade et al., 2015). Our results are also consistent with those of two recent studies using an optical transmission approach, which found that long dehydrating times (up to 70 h)  and very negative water potentials below the turgor loss point were necessary before vein embolisms were observed in leaf veins (Brodribb et al., 2016a(Brodribb et al., , 2016b. One of those studies showed a correlation between vein embolism and K leaf decline in four species (Brodribb et al., 2016b), although this was not necessarily causative, as K leaf appeared to decline by up to 50% before the turgor loss point and before any signal of embolism in leaf veins. Additionally, the sensitivity of K ox and K leaf may have been stronger under high irradiance. In that study, leaves were acclimated under low irradiance (less than 100 mmol quanta m 22 s 21 ). For many species, K leaf in hydrated leaves can be enhanced by many fold under high irradiance likely due to aquaporin expression (Cochard et al., 2007;Scoffoni et al., 2008;Maurel et al., 2015) and such high-light-acclimated leaves show stronger vulnerability before the turgor loss point (Guyot et al., 2012;Sack et al., 2016b). Similarly, a recent study partitioning the vulnerabilities of K x and K ox found that K ox was the strongest determinant of K leaf decline in two of four species (Trifiló et al., 2016), and, for the other two species, both xylem and outsidexylem pathways appeared to be strong drivers of K leaf decline. However, hydraulics measurements were performed in that study under low light, likely minimizing the response of K ox before the turgor loss point.
Our results for angiosperm leaves with their complex venation may be general for a yet greater diversity of plants, as two recent studies using microCT on needles of Pinus pinaster found few embolized conduits at needle water potentials that induced strong declines in K leaf (Charra-Vaskou et al., 2012;Bouche et al., 2016).
These findings suggest that the leaf outside-xylem pathways, in addition to experiencing the most negative water potentials in the plant, also have very strong hydraulic vulnerability. Such results are consistent with the hypothesis that strong K ox declines would act as a protective bottleneck, shielding the leaf and stem xylem under many scenarios of atmospheric and soil drought from tensions that would induce catastrophic embolisms (Scoffoni et al., 2014). Additional mechanisms for protection may operate; a recent study found that minor vein collapse in leaves of red oak (Quercus rubra) occurred under very strong tensions below the turgor loss point (more negative than 23 MPa) and, thus, could act as a further buffer against embolism under prolonged drought (Zhang et al., 2016). Notably, a similar protection occurs in roots, as cortical lacunae formation in fine roots induced strong declines in hydraulic conductance protecting root xylem conduits from embolism formation . Such a strong role of outside-xylem pathways in hydraulic decline in both leaves and roots suggests a general advantage throughout the plant of sensitive living tissues protecting the xylem from catastrophic embolism. Given that stem embolism may be in many or most cases irreversible (Urli et al., 2013), such a protective effect would be most important for longlived leaves and stems with high carbon investment, as commonly found in many drought-prone systems such as chaparral communities. This hypothesis of the importance of the K ox response was supported by our model simulations showing that whole-plant hydraulic conductance would decline under increasing soil drought and/or atmospheric drought (i.e. high VPD) primarily as a consequence of the strong declines in K ox . Because the leaves experience the lowest water potentials, and declining K ox provides an increasing bottleneck in the system, the tensions developed in leaf and stem xylem were, in most modeled scenarios, insufficient Neither leaf xylem hydraulic conductance (x; dashed light blue lines) nor stem xylem hydraulic conductance (s; dotted dark blue lines) experiences strong declines with increasing soil drought or VPD. The root hydraulic conductance (dashed red lines) declines strongly under increasing soil drought and to a smaller extent under increasing VPD. Because the model simulates a transpiring plant, when the soil water potential is at zero on the x axis, the transpiring leaf water potential is still substantially negative, driving the decline of K leaf from its maximum value (although not of K x ; for water potentials of each compartment, see Supplemental Table S2). Under the soil drought scenario, VPD was maintained at 0.5 kPa. Under the atmospheric drought scenario, soil water potential was maintained at 20.1 MPa.
to cause catastrophic embolism. The declines in K ox and K leaf may further protect the stem xylem from strong tensions and embolism if the strongly declining water potentials in the mesophyll influence stomatal closure, which tends to begin well above C TLP (Bartlett et al., 2016), and K ox could play an important role in stomatal control. Another potential advantage of outside-xylem pathways being more sensitive to dehydration is that they might recover more rapidly with water potential than embolized conduits in the xylem. Thus, changes in outside-xylem pathways with dehydration could be more reversible during drought and recovery cycles than xylem embolism. While xylem embolism requires several hours under no tension to recover by capillarity (Hochberg et al., 2016;Knipfer et al., 2016), in some species, K leaf can partially recover after only 1 h of rehydration , which could be due to the recovery of K ox . Future work should resolve the influence of K x and K ox decline on stomatal conductance and their recovery.
These results provide strong evidence for the role of outside-xylem pathways in driving changes in K leaf and whole-plant conductance under the range of water potentials that plants experience through mild and moderate drought stress. In contrast, after stomatal closure and under conditions of prolonged drought, sustained dehydration will induce embolism in leaf veins and, likely, in the stem xylem, eventually contributing to hydraulic failure and plant death (Anderegg et al., 2015).

Determinants of K ox Decline with Dehydration
Given the key role of K ox decline in dehydrating leaves, resolving the underlying causes is crucial. Experimental investigation remains challenging not only because of the complexity of liquid water movement through the living tissues outside the vein xylem but also because vapor-phase pathways contribute to K ox and thus K leaf (Pieruschka et al., 2010;Rockwell et al., 2014;Buckley et al., 2015). We implemented a spatially explicit model for the anatomical and biophysical determination of K ox (MOFLO; Buckley et al., 2015) and parameterized the model with our measurements of tissue structure in dehydrating leaves. These simulations showed that shrinking cells and air spaces in dehydrating leaves would in fact act to increase K ox due to the effects of shorter path lengths for water transport to the stomata, both horizontally, as effective vein length per leaf area increases, and vertically, from vein to stomata, given the shrinkage of the leaf thickness. Simulations showed that declines in membrane permeability could be important determinants of K ox decline that would drive K leaf decline overall, despite the effect of reduced tissue dimensions. A decline in membrane permeability could result from reduced aquaporin activity as cells dehydrate, a response demonstrated previously in studies using mutants of the model species Arabidopsis (Arabidopsis thaliana) and in cell probe studies of maize (Zea mays; Johansson et al., 1998;Kim and Steudle, 2007;Maurel et al., 2015). Furthermore, previous studies have found that aquaporin mutants and leaves of species previously perfused with aquaporin inhibitors to exhibit up to a 75% decrease of K leaf (Shatil-Cohen et al., 2011;Pou et al., 2013;Sade et al., 2015). Our findings are in line with the hypothesis that reduced aquaporin activity, potentially triggered by turgor decline and/or abscisic acid production during dehydration, would drive K ox decline (Shatil-Cohen et al., 2011) and further suggest that such a response would scale up to determine the decline of K leaf and whole-plant hydraulic decline. We found that, to model the observed declines of K ox due to the reduction of membrane permeability, it was necessary to posit an apoplastic barrier at the bundle sheath, analogous to the Casparian strip in root endodermis (Canny, 1986(Canny, , 1988, to constrain all water to exit the veins via bundle sheath cell membranes rather than via the apoplast. Such an apoplastic barrier has been supported previously by dye experiments (Shatil-Cohen et al., 2011;Shatil-Cohen and Moshelion, 2012) and hydraulics measurements in other species (Sack et al., 2004;Sade et al., 2014) and visualized in anatomical studies of some, but not all, species tested (Canny, 1986;Figure 6. Testing hypotheses for the potential drivers of the decline in K ox in dehydrating leaves, using a spatially explicit model of leaf outside-xylem water transport (see "Materials and Methods"). Parameterizing the model for four species, we estimated the K ox based on the decline of observed cell size, porosity (air space), and leaf area at the turgor loss point (light gray bars). Because in some cases these changes in tissue dimensions resulted in an increase in K ox , we modeled K ox decline according to three scenarios (always including the observed changes in tissue dimension): an 80% decline at the turgor loss point in membrane permeability (blue bars), cell connectivity (red bars), and cell wall thickness (dark gray bars). All simulations were run with or without an apoplastic barrier at the bundle sheath cells (solid versus striped bars). The yellow star on the x axis represents the observed percentage K ox decline at the turgor loss point. Across all four species, only simulations of a strong decrease in membrane permeability in leaves with an apoplastic barrier could explain the observed declines in K ox . Lersten, 1997;Wu et al., 2005;Ribeiro et al., 2007;Taneda et al., 2016). The restriction of water movement needed to explain declines in K ox could occur at the site at which water exits the vascular parenchyma to reach bundle sheath cells or cell walls or via a forced symplastic flow path through the vascular parenchyma cells until it reaches the bundle sheath; any of these mechanisms would strongly increase the resistance in water movement (Buckley, 2015). Elucidating whether such apoplastic barriers or symplastic flows through vascular parenchyma are typical is an important topic for future studies. Finally, modeling showed that changes in cell wall thickness during dehydration could strongly influence K ox (Fig. 6), given the important contribution of apoplastic cell wall pathways through the mesophyll in determining K ox at full hydration (Buckley, 2015). However, to our knowledge, such putative changes in cell wall thickness with dehydration have never been documented. Our results, along with the numerous aquaporin studies (see refs. above), most strongly support changes in membrane permeability at the vascular parenchyma or bundle sheath cell level as a mechanism for the decline in K ox with dehydration.
K ox Vulnerability in Relation to Dought Tolerance Across species, K ox and K x vulnerability during leaf dehydration correlated strongly with C TLP . C TLP is a good indicator of species drought tolerance across ecosystems, with more negative values present in species occurring in drier habitats or ecosystems (Bartlett et al., 2012). Recently, several studies have shown strong correlation of P 50,Kleaf with C TLP across diverse angiosperm species (Blackman et al., 2010;Scoffoni et al., 2012). These studies hypothesized that cells maintaining turgor at more negative water potentials could preserve cell integrity and, thus, hydraulic pathways outside the xylem and, therefore, confer resistance to hydraulic decline. However, given that our model simulations revealed that cell shrinkage would not cause a decline in K ox as hypothesized previously (Scoffoni et al., 2014), an indirect mechanism must underlie this correlation; for instance, a more negative C TLP may correspond to a greater ability to maintain cell membrane permeability, especially in the vascular parenchyma and/or bundle sheath (Kim and Steudle, 2007). The hypothesis that cell turgor loss might trigger aquaporin deactivation and/or abscisic acid production (Pierce and Raschke, 1980;Shatil-Cohen et al., 2011), which in turn would reduce membrane permeability, is consistent with recent work on cells and tissues in a range of species (Wan et al., 2004;Ye et al., 2005;Kim and Steudle, 2007;Shatil-Cohen et al., 2011;Brodribb and McAdam, 2013;Chaumont and Tyerman, 2014;McAdam and Brodribb, 2014;Vandeleur et al., 2014). Another source of the coordination of C TLP with the hydraulic vulnerability of the leaf and its compartments is that all of these physiologically important traits are coselected in species with greater drought tolerance (Blackman et al., 2010(Blackman et al., , 2014Bartlett et al., 2012Bartlett et al., , 2016.

CONCLUSION
Combining empirical, visual, and modeling approaches, we found that, in eight diverse species, the observed decline in K leaf during mild dehydration results primarily from losses in hydraulic conductance outside the vascular system (more than 75% across leaf dehydration from mild to extreme; 96% on average). These results indicate that outside-xylem processes are the main determinants of K leaf vulnerability to dehydration. Leaves avoid catastrophic xylem failure by regulating their K ox . After stomatal closure and under extreme drought, leaf vein and stem embolism might be unavoidable and induce catastrophic hydraulic failure. These findings pinpoint the mesophyll tissues, including the bundle sheath, as a central locus for the control of leaf and plant water transport during progressive drought.

Plant Material
Measurements were obtained for eight species diverse in phylogeny, origin, drought tolerance, and life form (Table I), growing in and around the campus of the University of California, Los Angeles, and Will Rogers State Park. Measurements were conducted from November 2013 to November 2014. The day before any of the measurements described below, shoots with a minimum of three nodes of stem below the leaves to be studied were excised in air from at least three individuals and transported in dark plastic bags filled with wet paper towels, where the shoot was recut underwater by a minimum of two nodes from the base and left to rehydrate overnight. We note that, although obtained in different years, both leaf and xylem hydraulic vulnerability curves were obtained from the same individuals, and no differences were found in K leaf values across years (Scoffoni et al., 2011;Guyot et al., 2012).

MicroCT
To directly visualize embolism in the xylem and structural changes in all orders of veins and in the mesophyll tissues, we used high-energy, highresolution microCT at the synchrotron at the Advanced Light Source in Berkeley, California (Beamline 8.3.2), in November 2014. Stacks of images were obtained by scanning the center (including the midvein) of living leaves on dehydrating shoots for four of our study species (Comarostaphylis diversifolia, Hedera canariensis, Lantana camara, and Magnolia grandiflora). Species were chosen for microCT based on their wide range of drought tolerance. A detailed description of sample preparation for microCT imaging is given in Supplemental Materials and Methods S1. Nine to 12 scans of the midrib and surrounding mesophyll at the center of the leaf were made per species of leaves spanning the whole range of leaf water potential obtained in the K x vulnerability curves (described below).
On three cross-sectional images randomly selected at the bottom, middle, and top part along the main axis of the microCT scan, conduit embolism in the midrib, along with mesophyll cell and tissue dimensions, were quantified. For each image, we measured the number of embolized conduits in the midrib and averaged for three areas of the leaf lamina measurements of the dimensions of tissues and cells (epidermis and cuticle, palisade mesophyll, spongy mesophyll, and palisade cell area, height, and diameter) using ImageJ software (version 1.46r; National Institutes of Health). Bundle sheath thickness and cell dimensions could not be resolved in these images. Three-dimensional volume renderings of our scans were made using Avizo 8.1.1 software (VSG) and used to determine the vein orders (identified by following the branching pattern from the secondary veins), and cross-sectional images at the start, middle, and end of the scanned region were used to determine the number of embolized conduits.
We calculated the %EMC at given leaf water potentials. Embolized conduits appear brightly in the images, but nonembolized conduits cannot be distinguished from each other or counted. Thus, we estimated the total number of midrib conduits in cross sections of these leaves using data taken from cross sections of three leaves sampled from the same plants of each species and visualized by light microscopy (Supplemental Fig. S3; for methods, see "Light Microscopy of Cells and Tissues within Leaves" below). Given that the number of midrib xylem conduits scales with the midrib vascular cross-sectional area for well-hydrated leaves of given species (Coomes et al., 2008;Taneda and Terashima, 2012), we counted the total number of xylem conduits in the midrib cross sections obtained from light microscopy for hydrated leaves and normalized by their midrib vascular area. These were averaged for each species to determine the conduit number per vascular area for hydrated leaves (CNA hydr ). Cross sections for both light microscopy and microCT scans were taken at the leaf midrib center. To calculate the total number of midrib conduits in cross sections of the scanned dehydrated leaves (CN), we had to account for the shrinkage of the midrib vascular area with water potential. For the scanned dehydrated leaves of each species, we plotted midrib vascular area for the dehydrated leaves (A dehyr ) and for the three fully hydrated leaves measured using light microscopy against leaf water potential (Supplemental Fig. S4) and thus estimated the proportion of area shrinkage relative to the value extrapolated to 0 MPa for each leaf (AS dehydr ). The conduit number (CN) for each individual scanned leaf was obtained as: We counted the number of embolized conduits in each scanned leaf (CN emb ) and calculated % EMC as: We note that the %EMC values differ slightly from those reported previously for the same images (Scoffoni et al., 2017), as we improved the calculation by adding the areas of the three light microscopy images of fully hydrated leaves to the regression against water potential to determine AS dehydr . This improved calculation resulted in no major changes in the patterns observed.
We considered the potential concern that the x-ray beam might produce damage artifacts that might have contributed uncertainty to the interpretation of the images. However, no damage from the x-ray beam was observed in our samples. Only a few gas-filled conduits were found at high water potentials in two species, which was to be expected given our sampling design (i.e. excising small shoots in air), as a small portion of conduits originating in the stem would extend into the leaf . Another indication that microCT faithfully represents mesophyll structure is that cell dimensions measured in the microCT scan images for hydrated leaves were statistically similar to those made on fully hydrated leaves of the same species using light microscopy (repeated-measures ANOVA was performed in Minitab 16; results are given in Supplemental Table S3).

Measuring Leaf and Leaf Xylem Hydraulic Vulnerability Curves
Leaf hydraulic vulnerability curves for seven of the eight study species were published previously for the same individuals used in this study (Scoffoni et al., 2011;Scoffoni and Sack, 2015), and that for Malosma laurina was constructed for this study. Measurements of K leaf vulnerability were made using the evaporative flux method (Supplemental Materials and Methods S1; Sack et al., 2002;Scoffoni et al., 2012), for which detailed protocols are available . All measurements were performed on leaves acclimated to high light for over 30 min (greater than 1,000 mmol photons m 22 s 21 ). We constructed K x vulnerability curves using the vacuum pump method (Supplemental Materials and Methods S1) for the same individuals and species from which K leaf vulnerability curves were obtained. Data for four species (i.e. Comarostaphylis diversifolia, Hedera canariensis, Quercus agrifolia, and Salvia canariensis) were published previously in a study of potential methodological artifacts in leaf hydraulic measurements , and additional measurements were made here for four other species (Cercocarpus betuloides, Lantana camara, Magnolia grandiflora, and M. laurina).
To construct hydraulic vulnerability curves, we selected the maximum likelihood function that best fitted data for each species using the optim function in R 3.1.0 (http://www.r-project.org; Burnham and Anderson, 2004;Scoffoni et al., 2012). Five functions were tested according to previous studies (Pammenter and Vander Willigen, 1998;Scoffoni et al., 2012): a linear function (K z ¼ aC z þ b), a two-parameter sigmoidal function (K z ¼ 100 1þ e ðaðCz 2 bÞ ), a three-parameter sigmoidal function (K z ¼ a 1þe 2 ð Cz 2 x 0 b Þ ), a logistic function , and an exponential function (K z ¼ y 0 þ ae 2 bCz ). The K z and C z in the above functions represent either the K leaf or K x and water potentials. Functions were compared using the Akaike Information Criterion (AIC) corrected for low n. The function with the lowest AIC value (differences of greater than 2 considered) was chosen as the maximum likelihood function.

Determination of Leaf Outside-Xylem Vulnerability Curves
Based on Equation 1, we constructed K ox vulnerability curves from K leaf and K x values along the water potential range tested for given species (i.e. from maximum K leaf until it had declined to a negligible level). Thus, for the different water potentials, each K ox point was obtained as the reciprocal of the difference between K leaf 21 and K x 21 following Equation 1. For background and justification of this subtraction method, see Supplemental Materials and Methods S1.

Whole-Plant Hydraulic Model Simulations
We modeled the influence of leaf hydraulic declines on the plant hydraulic system under simulated soil and atmospheric drought using a previously described approach (Osborne and Sack, 2012). The plant hydraulic stomatal model is based on Darcy's law, assumes steady-state flow, and simultaneously resolves water potentials and hydraulic conductance for each plant component, given inputs of soil water potential and VPD and parameters for the response of the hydraulic conductance of whole root, whole stem, leaf xylem and outside xylem, and stomatal conductance to water potential within the respective organ. For the four species tested, we simulated the impact of declining soil water potential or increasing VPD given the measured vulnerability curves for K ox and K x , obtained as described above. We did not have data for the response of the stem, root, or stomata to dehydration for these species, so we used estimates based on current understanding in the literature. Thus, we assumed the vulnerability curve of the whole-stem xylem to follow a sigmoid pattern, with maximum hydraulic conductance representing half of the whole-plant resistance (Tyree and Zimmermann, 2002). To be conservative, we assigned to the stem a water potential at 50% loss of hydraulic conductance equal to that of the leaf xylem, since xylem conduits in the stem are expected to undergo air seeding at similar or more negative water potentials (Tyree and Ewers, 1991;Choat et al., 2005). Thus, the stem xylem was modeled as potentially more sensitive than it might be in reality, making more robust our finding of its low hydraulic decline when the whole plant is droughted, due to the role of leaf hydraulic decline in minimizing tensions in the stem. We assumed the root vulnerability curve to be equal to the whole-leaf hydraulic vulnerability curve (obtained as described above), given that, on average, the root and leaf contribute approximately the same resistance throughout the whole plant (Tyree and Zimmermann, 2002) and have both xylem and extraxylem pathways for water movement (Tyree and Zimmermann, 2002). We set the stomatal conductance decline with leaf water potential as similar to that of the vulnerability of the leaf outside-xylem pathways, using a maximum stomatal conductance value of 300 mmol m 22 s 21 across species. A range of alternative parameterizations did not change the overall findings (data not shown). We note that future work will enable more precise calibration of the model (e.g. with vulnerability functions for all organs). Simulations were run in Python 2.7.10 using the future, scipy, and pandas packages. Model code is available on request.

Modeling the Outside-Xylem Flow Pathways with Dehydration
We used a spatially explicit model of outside-xylem flow pathways in the leaf (MOFLO; Buckley et al., 2015) that can be parameterized with leaf anatomy to investigate potential causes of the strong declines in K ox observed with dehydration. We first simulated the impact of anatomical changes alone, based on anatomical measurements at different water potentials, including epidermal, spongy, and palisade mesophyll cell shrinkage (obtained from micro-CT images as described above; Fig. 5), percentage leaf area shrinkage (which influences vein length per leaf area), and percentage intercellular air space change (published previously for these same species and individuals; Scoffoni et al., 2014). Since bundle sheath cell area could not be determined in the micro-CT images, we assumed that these cells shrank by the same percentage as spongy mesophyll cells. We then simulated the impact on K ox of the decline in membrane permeability, cell connectivity, and cell wall thickness at the turgor loss point, using values for tissue dimensions observed at the turgor loss point. Given that we did not have measurements of membrane permeability, cell connectivity, and cell wall thickness at the turgor loss point, we estimated the reduction in these parameters required to cause the observed decline in K ox at the turgor loss point. We repeated all of these simulations under two scenarios: with and without an apoplastic barrier at the bundle sheath cells.

Measurement of the Turgor Loss Point
The leaf turgor loss point for seven of eight species was obtained from pressure-volume curves of previously published studies (Scoffoni et al., , 2014 that were based on the same individuals of the study species. Pressurevolume curves were obtained for five leaves of three individuals of M. laurina in the fall of 2014 using a detailed published standard protocol (Sack, 2010).

Light Microscopy of Cells and Tissues within Leaves
For measurements of leaf cross-sectional anatomy, we used images from a previously published study of different anatomical traits made on the same individuals of four study species (John et al., 2013). Briefly, from each leaf center, a 1-3 0.5-cm rectangle was cut and embedded gradually in low-viscosity acrylic resin (L.R. White; London Resin) in ethanol, under vacuum over the course of 1 week, then dried at 55°C overnight. Samples were then sectioned using glass knives (cut using an LKB 7800 KnifeMaker; LKB Produkter) at 1 mm thickness in a rotary microtome (Leica Ultracut E; Reichter-Jung). Sections were stained in 0.01% Toluidine Blue in 1% sodium borate and imaged using 53, 103, 203, and 403 objectives using a light microscope (Leica Lietz DMRB; Leica Microsystems) with a camera utilizing SPOT advanced imaging software (SPOT Imaging Solutions; Diagnostic Instruments) for a total image magnification of 2873 to 2,3003. Using ImageJ, we measured the vascular bundle area in the midrib and counted the total number of xylem conduits.

Statistics
To test the causal influences of xylem and outside-xylem conductance decline on whole-leaf hydraulic decline, we used three analyses. First, we calculated causal effects within species by partitioning changes in leaf resistance (R leaf = 1/K leaf ) into changes in xylem resistance (R x = 1/K x ) and outsidexylem resistance (R x = 1/K ox ); since R leaf = R x + R ox , DR leaf = DR x + DR ox , where D denotes a change between full turgor and either the turgor loss point or P 50 . Thus, for example, the percentage of leaf hydraulic decline due to outsidexylem pathways was calculated as DR ox /DR leaf 3 100%. Then, we estimated the importance of K x and K ox decline in explaining species differences in leaf hydraulic vulnerability (i.e. in P 50,Kleaf ). We tested whether P 50,Kleaf was best predicted by the water potential at 50% decline in xylem hydraulic conductance (P 50,Kx ) or that of outside-xylem hydraulic conductance (P 50,Kox ), or their combined effect, according to the following models: P 50,Kleaf = a+bP 50,Kx , P 50, Kleaf = a+bP 50,Kox , or P 50,Kleaf = a+bP 50,Kox + cP 50,Kx . We used maximum likelihood selection of the best model using the optim function in R 3.1.0 (Burnham and Anderson, 2004;Scoffoni et al., 2012). The model with the lowest AIC corrected for low n by at least 2 was selected as the maximum likelihood model. We also applied independent effects analysis, which is suited to robustly determine the contribution of correlated predictor variables to an output variable (Murray and Conner, 2009), and thereby calculated the percentage contribution of P 50,Kx and P 50,Kox to the variation across species in P 50,Kleaf , using the hier.part function in R.3.1.0.

Supplemental Data
The following supplemental materials are available.
Supplemental Figure S1. Decline of leaf K x with dehydration.
Supplemental Figure S2. Model simulations of plant hydraulic response to dehydrating soil and increasing VPD for four diverse species.
Supplemental Figure S3. Light microscopy midrib cross sections of the four study species used for microCT.
Supplemental Figure S4. Percentage midrib vascular area of maximum at full hydration plotted against leaf water potential.
Supplemental Table S1. Parameters for the three models tested to best predict P 50,Kleaf .
Supplemental Table S2. Inputs and results for the whole-plant hydraulic model simulations.
Supplemental Table S3. Means 6 SE of cell dimensions measured from microCT scans and light microscopy.
Supplemental Materials and Methods S1.