Sap flow through petioles and petiolules reveals leaf-level responses to light and vapor pressure deficit in the tropical tree Tabebuia rosea (Bignoniaceae)

Continuous measurements of sap flow have been widely used to measure water flux through tree stems and branches. However, these measurements lack the resolution necessary for determining fine-scale, leaf-level responses to environmental variables. We used the heat ratio method to measure sap flow rates through leaf petioles and leaflet petiolules of saplings of the tropical tree Tabebuia rosea (Bignoniaceae) to determine how leaf and leaflet sap flow responds to variation in light and vapor pressure deficit (VPD). We found that in the morning sap flow rates to east-facing leaves increased 26 minutes before adjacent west-facing leaves. Although leaves had higher integrated sap flow than their largest leaflet, this difference was not proportional to the difference in leaf area, which could be due to lower conduit area in petiolules than in petioles. In contrast to measurements on main stems, integrated daily sap flow was negatively correlated with daily mean VPD. Furthermore, leaves exhibited previously undescribed patterns of hysteresis in the sap flow-VPD and sap flow-PAR relationships. When hysteresis in the sap flow-PAR relationship was clockwise, the sap flow-VPD relationship was also clockwise; however, when hysteresis in the sap flow-PAR relationship was counterclockwise, the sap flow-VPD relationship displayed an intersected loop. These pattern differences highlight how substantially leaf-level processes may vary within a canopy and how leaf-level processes may not scale predictably to the stem level.


INTRODUCTION
Approximately 90% of all water converted from the liquid phase to the vapor phase in terrestrial ecosystems moves through plants (Jasechko et al. 2013); in the tropics this amounts to an estimated 32x10 15 kg of water per year (Hetherington and Woodward 2003). Almost all of this water transits through plant leaves. Understanding how leaves respond to abiotic drivers is important for modeling efforts at scales from leaves to landscapes (Jarvis and McNaughton 1986). At the leaf level, knowing which drivers impact transpiration most under daily and seasonally varying conditions is critical to understanding what may limit the distribution and abundance of species across the globe.
Various sap fow methods are commonly used to estimate almost continuously tree responses to environmental conditions for extended time periods (Marshall 1958, estimates of whole-tree water use (Wullschleger et al. 1998), their utility for describing leaf-level processes can be limited by a variety of factors including time lags, capacitance, hydraulic resistance, and variation in these factors along the root-to-leaf continuum.
Rarely have researchers attempted to measure sap fow rates through petioles of individual leaves (Sheriff 1972). Recently, Clearwater (2009) (Roddy and Dawson 2012, Goldsmith et al. 2013) and through stems of anatomically and phylogenetically diverse species of the South African fynbos fora (Skelton et al. 2013). These studies show that measuring sap fow directly adjacent to transpiring leaves can deepen our understanding of how leaves respond to variation in environmental conditions across a range of timescales. Placing sensors in close proximity to the transpiring leaves has the advantage of fne-scale measurements akin to leaf gas exchange without the disadvantage of enclosing leaves in a cuvette that removes the leaf boundary layer and otherwise modifes the leaf microenvironment.
Variation in sap fux is infuenced by a variety of environmental conditions, including soil water availability, vapor pressure defcit (VPD), and solar radiation, and diurnal patterns may also vary seasonally (e.g. O'Grady et al. 1999, Zeppel et al. 2004). Over diurnal cycles, a change in an environmental variable in the morning does not always produce an equivalent response in sap fow as it does in the afternoon.
Such a pattern is termed hysteresis and has been commonly observed in the sap fow responses to light and VPD. For example, at a given VPD that occurs both in the morning and again in the afternoon, sap velocity is higher in the morning (when VPD is increasing) than in the afternoon (when VPD is decreasing), creating a clockwise different stomatal responses to light and to VPD. Because VPD reaches its daily peak a few hours after light reaches its daily maximum at solar noon, sap velocity will be higher in the afternoon, when VPD is higher and stomata are fully open. Zeppel et al. (2004) argue that stomatal conductance saturates at relatively low light levels in the morning, and that above this saturating light level, VPD becomes the predominant driver of transpiration and sap fow. These responses probably vary between leaves acclimated to different microenvironments (e.g. between sun-and shade-leaves).
Using sap fow measurements on main stems of canopy trees to test these hypotheses for the causes of hysteresis are thus fraught with potential problems that focusing on leaves may circumvent. For leaves, sap fow responses to VPD are often similar to those for stems, although under some conditions leaves show different patterns (Roddy and Dawson 2013). In addition to clockwise hysteresis in the responses to VPD, sap fow through petioles sometimes exhibits an intersected loop (or 'fgure-eight') pattern in response to diurnal variation in VPD. Determining the sap fow responses to environmental variables of individual leaves provides an opportunity to better elucidate important dynamics of plant water use. Furthermore, incorporating explicit measurements of sap fow to individual leaves could help to improve upon methods for scaling up to whole canopy processes.
In the present study, we measured sap fow rates through petioles and petiolules of saplings of the tropical tree Tabebuia rosea (Bignoniaceae) to understand how sap fow responds to variation in light and VPD. Because the fgure-eight pattern of hysteresis in the sap fow-VPD relationship reported by Roddy and Dawson (2013) may result from an interaction with light, we also measured photosynthetically active radiation (PAR) levels on each leaf or leafet to determine the conditions under which different patterns of hysteresis may occur. We were particularly interested in examining the differences in sap fow patterns between adjacent leaves and between leaves and leafets because different microenvironmental conditions may cause sap fow patterns to differ between leaves on the same stem. Furthermore, differences in sap fux through petioles and petiolules may refect variation in hydraulic architecture. If the hydraulic pathway constricts downstream, then sap velocity must increase as it moves towards leafets. Our results highlight how measuring sap fow rates to individual leaves could deepen our understanding of the linkages between hydraulic architecture and plant water use.

Plant Material
Tabebuia rosea (Bignoniaceae) grows to become a canopy tree in the lowland forests of central Panama. While adults are often deciduous, seedlings are evergreen with fve palmate leafets of varying size encircling the petiole. Plants were grown from seed in 20-liter, insulated pots outdoors under a glass roof at the plant growth facilities of the Smithsonian Tropical Research Institute in Gamboa, Panama, until a few days before sap fow measurements were begun. When the tops of the plants were ~70 cm above the soil surface, they were transferred to a glass chamber to protect them from strong afternoon winds. The lower ~1 m of this chamber was made of cement painted white, and doors on east-and west-facing sides of the chamber were left ajar to allow air circulation. At the beginning of the sap fow measurements, the tops of the plants were even with the top of the cement wall at the base of the chamber, and during the course of the experiment two new sets of leaves were produced. Pots were kept wellwatered except for one week when water was withheld to determine how sap fow rates would respond to declining soil water. This week coincided with a dramatic increase in VPD. Sensors were installed when plants were approximately eight months old, a few days after transferring them to the glass chamber. On each measured leaf, sap fow sensors were installed on the leaf petiole and on the petiolule of the middle, largest leafet. On each plant, two adjacent leaves on opposite sides of the plant were chosen for measurement. At the time of installation, these leaves were the newest, fully-expanded leaves on each plant. Plants were positioned so that the axis defned by the two measured leaves on each plant were oriented east-west. Of the total 12 sensors installed, fve failed, leaving two sensors on petiolules and fve sensors on petioles.
Sap fow sensors and measurements were based on the design and theory of Clearwater (2009) with some slight modifcations described previously Dawson 2012, 2013) and again briefy here. Sensors were constructed from a silicone backing and were connected to 10 cm leads with Molex connectors that were then connected by 10 m leads to an AM16/32 multiplexer and CR23X datalogger (Campbell Scientifc Inc., Logan, UT). Sensors were held in place with paraflm, and sensors and connections were insulated with multiple layers of bubblewrap and aluminum foil at least 2 cm above and below the sensor. Consistent with previous applications of the HRM, we measured initial temperatures for 10 seconds prior to fring a 4-second heat pulse, monitored temperatures every 2 seconds for 200 seconds after the heat pulse, and initiated the measurement routine every 10 minutes.
The heat pulse velocity, v h (cm s -1 ), was calculated from the temperature ratio as: where v h is the heat pulse velocity in cm s -1 , k is the thermal diffusivity (cm 2 s -1 ), x is the distance from the heater to each of the thermocouples (cm), and δT 1 and δT 2 are the temperature rises ( o C) above and below the heater, respectively (Marshall 1958, Burgess et al. 2001, Clearwater et al. 2009). We estimated the thermal diffusivity as: where t m is the time (seconds) between the heat pulse and the maximum temperature rise recorded x cm above or below the heater under conditions of zero sap fow (Clearwater et al. 2009). We measured t m every morning before dawn when atmospheric vapor pressures were lowest (between 0500 and 0600 hrs). At this time, the vapor pressure defcit was almost always below 0.3 kPa, and therefore we assumed v h was approximately zero. Thermal diffusivity, k, was calculated for each thermocouple (upstream and downstream) from these predawn measurements of t m , averaged for each sensor, and used to calculate v h from the heat ratios for the subsequent 24 hours.
Measurements of k on nights with VPD always above 0.3 kPa were discarded and replaced with the most recently measured k when VPD < 0.3 kPa. We estimated the temperature ratio under zero-fow conditions by excising petioles and petiolules above and below the sensor at predawn at the end of the experiment, greasing the cut ends, placing the segments in a darkened box, and recording the temperature ratios for the subsequent ~4 hours. The average of these zero-fow temperature ratios corresponded very well with the temperature ratios recorded predawn under low VPD (less than ~0.3 kPa) conditions. The sensor-specifc average temperature ratio under zero-fow conditions was subtracted from all calculated heat ratios. This corrected heat ratio was then used to calculate v h .

Measurements of light and vapor pressure defcit
Light measurements were made using S1787 photodiodes (Hamamatsu Photonics, Hamamatsu City, Japan). Photodiodes were connected to 15 cm long copper wires with Molex connectors and then to 10 m leads, which were connected to a CR5000 datalogger measuring in differential mode. Circuits created by each photodiode were closed with a 100 Ohm resistor. Photodiodes were installed just above each leafet with a sap fow sensor, and the photodiode was positioned to be parallel to the axis of the central vein of the leafet. Light measurements were made every minute and averaged Vapor pressure defcit was calculated from temperature and relative humidity measurements made every 10 minutes with a HOBO U23 datalogger (Onset Computer Corp., Bourne, MA) that was housed in a covered, white, PVC, Y-shaped tube and hung level with the tops of the plants.

Data analysis
All analyses were performed using R (R Core Team 2012). Raw velocity measurements were processed following previously published methods (Roddy and Dawson 2012, Skelton et al. 2013. Measurements of v h were smoothed using the 'loess' function, which fts a polynomial to a subset of the data in a moving window of 35 points. For analyses of structure (leaf vs. leafet) or aspect (east-vs. west-facing leaves), sap fow measurements from individual sensors in each group were averaged. To estimate the total sap fow during the day and night, we integrated the time course of vh measurements for each day and night using the 'auc' function in the package MESS, which calculates the area under the curve using the trapezoid rule. Daytime was defned as being between 600 and 1800 hours, which corresponded to morning and evening twilight. To minimize the effects of nocturnal reflling, we defned nighttime as being between 100 and 600 hours, which assumed that diurnal water potential declines had mostly recovered within seven hours after sunset. To analyze the effects of VPD on integrated sap fow rates, we linearly regressed integrated sap fow against mean VPD.
In all regressions for leaves and leafets in the day and in the night, the linear model was determined to be as good or better than both the logarithmic and power functions by comparing the residual standard errors.
Differences in the timing of morning sap fow between east-and west-facing leaves were compared at a critical v h of 1.5 cm hr -1 . We chose this critical value because it was higher than any measured nighttime velocities and lower than most daytime velocities. We estimated the time at which v h = 1.5 cm hr -1 by assuming a linear relationship (y = mx + b) between the two sequential morning measurements that spanned v h of 1.5 cm hr -1 . To compare east-versus west-facing leaves and leaves versus leafets, we used linear mixed effect models with day as the random variable, which accounts for repeated measures.

RESULTS
Daily maximum VPD varied from 2.8 kPa to 6.9 kPa during the experiment, while daily maximum PAR at the top of the canopy varied from 1125 to 1640 µmol m -2 sec -1 .
The daily maximum sap fow rate through petioles varied between 1.4 cm hr -1 to 4.5 cm hr -1 . This lowest daily maximum v h occurred at the end of a week without water, during which time fve of the seven hottest, driest days occurred. Nighttime v h through petioles varied throughout the study, but was always below 1.0 cm hr -1 and below 0.5 cm hr -1 on all but seven nights. Overall, thermal diffusivity, k, ranged from 0.00136 to 0.00170 cm 2 s -1 . There were slight differences in k between sensors, but k was relatively constant throughout the experiment for each sensor, consistent with previously reported values for k from a diverse set of plant structures and species (Clearwater et al. 2009, Roddy and Dawson 2012, Skelton et al. 2013).
On every morning, sap fow rates to east-facing leaves increased more quickly than did sap fow rates to west-facing leaves. East-facing leaves had sap fow rates of 1.5 cm hr -1 on average 26 minutes before sap fow rates to west-facing leaves reached the same threshold (t = 5.67, df = 23, P < 0.001; Figure 1). In addition, on 18 out of 25 days, west-facing leaves reached their daily peak sap fow rate later in the day than east-facing leaves. However, sap fow rates to east-facing leaves did not decline any earlier in the evening than west-facing leaves, and east-facing leaves generally had higher nighttime sap fow rates than west-facing leaves, perhaps indicative of greater reflling.
Patterns of sap fow to leaves and leafets also differed. Leafets generally had lower sap fow rates than leaves, and the daily integrated sap fow through petioles and petiolules differed signifcantly (t = 7.42, df = 24, P < 0.001; Figure 2). While water was withheld for one week, daily maximum sap velocities for both leaves and leafets declined such that leafet sap fow rates were about half of those to leaves (Figure 2a).
On the day immediately following re-watering, leaves and leafets had almost equivalent sap fow rates, which continued to increase on subsequent days despite declining daily maximum VPD during these days. Daytime integrated sap fow to leaves and leafets was negatively correlated with mean VPD (Figure 3), both when including all days and when the last fve days of the drought treatment were excluded (Table 1). There was a signifcant, negative relationship only between nighttime integrated sap fow of leafets and mean nighttime VPD, but only when all data, including the drought days, were included. There was no relationship between nighttime VPD and integrated sap fow for leaves. Maximum v h for leaves occurred at a slightly higher VPD than it did for leafets (2.21 kPa vs. 2.06 kPa; grey symbols in Figure 3).
Patterns of sap fow hysteresis can be grouped into two classes, exemplifed by data from two days from the same leaf shown in

DISCUSSION
Leaf physiology responds rapidly to changes in the leaf microenvironment in ways not previously appreciated (Zhang et al. 2013), and leaves may protect stems from low water potentials that can lead to loss of xylem functioning (Sperry 1986, Hao et al. 2008, Chen et al. 2009, Johnson et al. 2011, Bucci et al. 2012, Zhang et al. 2013). As a result, there has been burgeoning interest in the diurnal variability of leaf hydraulic functioning (Brodribb and Holbrook 2004, 2007, Johnson et al. 2009 increased, on average, 26 minutes earlier in the morning than they did to west-facing leaves, due to earlier increases in leaf-level PAR to east-facing leaves than to westfacing leaves. Despite these differences in timing, leaf hydraulic conductance of eastand west-facing leaves may be similar if higher transpiration rates in east-facing leaves are accompanied by greater declines in leaf water potential. Leaf microclimatic conditions can cause substantial differences between even adjacent leaves on the same branch (Figure 1), which can infuence the dynamics of branch sap fow (Burgess and Dawson 2008). How much infuence water use by one leaf may have on water use by another, adjacent leaf is likely related to xylem hydraulic architecture. In addition to fowing longitudinally from roots to leaves, water may also fow laterally within a stem (MacKay and Weatherley 1973, James et al. 2003, Schulte and Costa 2010. The degree of this lateral fow varies among species and results from lateral connections between adjacent xylem vessels. Highly sectored xylem leads to close coupling of water uptake by roots on one side of the plant and water use by leaves on the same side of the plant. In this case, adjacent leaves on different sides of the plants would draw upon largely different pools of water in the stem. High sectoriality in xylem architecture allows plant parts to function independently, such that branches or leaves may compete little for water (Brooks et al. 2003, Orians et al. 2005. In contrast, highly integrated xylem (low sectoriality) leads to tighter hydraulic linkages between adjacent leaves on different sides of the stem axis. While we do not know how well integrated the xylem of adjacent leaves in T. rosea may be, orthostichous leaves (vertically aligned along the shoot axis) generally have more interconnected vasculature than do nonorthostichous leaves (those on different sides of a shoot; Watson andCasper 1984, Orians et al. 2005). Thus, adjacent east-and west-facing T. rosea leaves probably function more independently than would two east-facing, orthostichous leaves. Sap fow to individual leaves varies among leaves on the same branch, and the magnitude of this variation may itself vary among species depending on xylem architecture.
Patterns of sap fow through petioles were similar to patterns observed for petioles of other tropical species Dawson 2012, 2013), but, in some cases, different from patterns observed for main stems. On some days, patterns of hysteresis in the v h -VPD relationship were similar to those seen for main stems of canopy trees (Meinzer et al. 1997, O'Grady et al. 1999, Zeppel et al. 2004; Figure 4a). In this type of hysteresis, v h was higher in the morning than in the afternoon for a given VPD, creating a clockwise loop in the relationship between v h and VPD. On days when the v h -VPD relationship showed a clockwise loop, the v h -PAR relationship also had a clockwise hysteresis loop (Figure 4b). In contrast, for main stems, a clockwise v h -VPD loop is normally accompanied by a counterclockwise v h -PAR loop (Zeppel et al. 2004). On days with this frst type of hysteresis, v h was higher in the morning than in the afternoon, with maximum daily v h occurring closer in time to peak PAR than to peak VPD. On these days, PAR peaked early in the day, saturating stomatal conductance and leading to high v h even when VPD was moderate. However, we observed a second type of hysteresis, characterized by a counterclockwise loop in the v h -PAR relationship ( Figure   4e) that, unlike for main stems, was accompanied by a markedly different relationship between v h and VPD. When the v h -PAR relationship exhibited counterclockwise hysteresis, the v h -VPD relationship was characterized by an intersected loop, or fgureeight ( Figure 4d). Although this intersected loop has been reported previously for main stems (Meinzer et al. 1999, O'Grady et al. 1999, its meaning has not been fully discussed or understood. This pattern occurred when afternoon v h was higher than morning v h , causing maximum daily v h to occur closer in time to peak daily VPD than to peak daily PAR. If morning transpiration is low and does not result in substantial water potential declines, then v h may peak in the afternoon when VPD peaks, as occurred on the second day shown. Why morning sap fow on this day was so low remains unclear but may be due to low water potentials, which we did not measure. Regardless, this second type of hysteresis exhibiting an intersected loop requires (1) Figure 4, midday depression alone may not lead to the intersected loop hysteresis. The second day did, however, have a higher afternoon VPD both in absolute terms (maximum of 4.77 compared to 3.44 kPa) and relative to PAR (Figure 4c,f), which was probably partly responsible for increased afternoon transpiration.
The most probable cause underlying such different patterns of hysteresis for individual leaves and for main stems is likely to be a matter of scale. Sap fow through stems integrates the individual sap fow responses of many leaves in drastically different microclimates. One of the most obvious sources of within-canopy variation is between different parts of a plant canopy that undergo different diurnal patterns of incident PAR, yet measurements on main stems ignore most of this within-canopy variation. In the present study, leaf aspect infuenced patterns of sap fow, and previous studies on branches have shown that aspect infuences both absolute rates of sap fow and the timing of peak sap fow within the day (Steinberg et al. 1990, Akilan et al. 1994, Martin et al. 2001, Alarcón et al. 2003, Burgess and Dawson 2008 Figure 1). Time lags between daily peaks of sap fow for east-and west-facing branches of large trees would result in different patterns of hysteresis depending on branch aspect (Burgess and Dawson 2008), and these patterns for branches may be similar to the second type of hysteresis (the fgure-eight) we report for individual leaves. By measuring incident PAR to each leaf, we attempted in the present study to account for some of the variation in leaf microclimate that infuences sap fow. However, we still ignored some important factors, such as leaf temperature and its effects on leaf saturation vapor pressure and the vapor pressure gradient (VPG) driving transpiration. This may be an acceptable oversight because atmospheric humidity has a greater impact on stomatal conductance than does leaf temperature (Fredeen andSage 1999, Mott andPeak 2010). In addition to microclimatic variation, leaves and stems differ in their hydraulic architecture, which could infuence sap fow patterns and hysteresis. Leaf water balance changes rapidly as effux and infux of water vary asynchronously on the timescale of seconds Sinclair 1973, Sheriff 1974). Water balance of stems may not change as rapidly, however, because of the compensatory effects of having numerous parallel pathways for water entry and loss. Thus, transpiration and sap fow may vary over much shorter timescales for leaves than for stems. Examining and quantifying sap fow hysteresis may provide new insights into hydraulic functioning in response to various abiotic factors infuencing transpiration (Zeppel et al. 2004, Pfautsch and Adams 2013, Roddy and Dawson 2013. Integrated daily plant water use, as measured by sap fow, generally increases with increasing mean and maximum daily VPD for plants from a wide variety of habitats, including canopy trees and shrubs (e.g. Zeppel et al. 2004, Pfautsch and Adams 2013, Skelton et al. 2013. However, in our experiment integrated daily leaf water use decreased with increasing mean daily VPD (Figure 3), whether days of declining soil water content were included in the analysis or not (Table 1). There was a signifcant negative relationship between integrated nocturnal sap fow and VPD for leafets, but not for leaves, although this relationship was driven by very low nighttime sap fow during the drought (Table 1). For both leaves and leafets, the VPD at which maximum daily v h occurred was remarkably well conserved across days (2.21 and 2.06 kPa, respectively) and was, interestingly, the same whether maximum v h occurred in the morning or in the afternoon (Figure 4a,d). These patterns opposite to those seen in main stems may result from higher than normal VPDs during our experiment. Leaves of T. rosea saplings may rarely encounter such high daytime VPD under natural conditions, and stomatal sensitivity to VPD may be responsible for the negative relationship we observed (Oren et al. 1999b). At VPD above ~2 kPa, instantaneous sap fow rates often declined, consistent with stomatal closure to regulate transpiration rate and leaf water potential. For T. rosea saplings, the VPD at maximum v h was higher than the VPD at maximum g s of other species, perhaps because of the higher than normal VPD during our experiment and the time lag between reaching maximum g s and maximum v h due to hydraulic resistance.
In response to declining soil water availability, daily maximum v h declined for both leaves and leafets despite increasing VPD during this time. Rewatering caused an immediate increase in leafet v h , such that it almost equaled leaf v h (Fig. 2).
Assuming the ratio of leafet area to conduit cross-sectional area (LA:SA ratio) is the same for all leafets, then instantaneous and integrated leafet sap fow, as a fraction of leaf sap fow, should be proportional to leafet area. However, both instantaneous and integrated leafet sap fow were higher than this prediction, probably because crosssectional conduit area of petiolules is lower than that of petioles. This could result from a combination of conduit taper and differences in the number of conduits between ranks (McCulloh et al. , 2010. Although we did not measure conduit dimensions, our results highlight the potential linkages between leaf hydraulic architecture and diurnal patterns of water use at different scales. As of yet, there has been remarkably little effort to connect xylem structure-function relationships to continuous, sap fow measurements of plant water use.         (c,f) However, the VPD-PAR relationship was approximately the same for the two days. Table 1. Summary statistics for the linear regressions between integrated sap distance and mean VPD for leaves and leafets in the day and in the night, whether including data from the week of drought or not.