Leaf water ¹⁸O and ²H maps show directional enrichment discrepancy in Colocasia esculenta

IM-CRDS analysis sequence

The IM-CRDS analysis sequence was adapted from a protocol developed by van Geldern & Barth (2012) for liquid water samples. Following their notation, Table 2 presents the sequence of standards and samples used in our study. Six empty vials were analysed at the beginning of each run. The average water vapour content, δ¹⁸O, and δ²H of the six vials were measured and introduced in a mixing model that allowed the signal from the ambient air present in the vial to be removed to retrieve the true isotope composition of the sample analysed:

(1)

where f_air = [H₂O]_air/[H₂O]_measured, with [H₂O]_air and [H₂O]_measured the water vapour concentrations inside the blank vial and the sample vial, respectively. Reference water samples (ABB – Los Gatos Research Inc., San Jose, CA, USA; working standards, all versus Vienna Standard Mean Ocean Water (VSMOW): ER4: δ¹⁸O = 81.33 ± 0.3‰, δ²H = 654.1 ± 1.0‰; ER1: δ¹⁸O = 12.34 ± 0.3‰, δ²H = 108.7 ± 1.0‰; 5A: δ¹⁸O = −2.80 ± 0.15‰, δ²H = −9.5 ± 0.5‰; and 3: δ¹⁸O = −11.54 ± 0.1‰, δ²H = −79.0 ± 1.0‰) were injected on punch holes of glass filter paper provided with the IM, and the same piece of filter paper was reused for all the injections of a single reference water. We found that 3μL of reference water was necessary to reproduce the peak of water vapour concentration (in the analyser) produced by one punch hole of C. esculenta (Cui et al. 2017).

Following the protocol developed for liquid water samples in previous work (van Geldern & Barth, 2012), the data were corrected for drift and memory effects and rescaled back to VSMOW (Table 2). The drift of the instrument was determined by rerunning the drift-monitoring standard (DEST, working standard ER1) at the beginning, the end and half-way through the samples. This standard water was chosen to have an isotope composition close to that of the leaf samples analysed. A linear regression was then used to realign all the measurements, according to the sample order. The effect of the drift was found to be small for both isotope ratios, with a slope of −0.004 ± 0.005 per sample in δ¹⁸O and −0.05 ± 0.07 per thousand per sample in δ²H. These values are comparable with values found for liquid water samples (van Geldern & Barth, 2012). For the memory effect, three standard waters with large isotopic differences were injected 10 times in a row each. An isotopically high (HIS, working standard ER4) and low (ANTA, working standard 3) standard waters were used along with DEST to determine memory effects from high-to-low and low-to-high transitions. A memory coefficient was associated with each successive injection. The coefficients were determined using the ‘Solver’ function in Excel (Microsoft, Redmond, WA, USA) so as to minimize the standard deviation of all three standard waters simultaneously. The memory coefficients are presented in Supporting Information Figs S1 and S2. The evolution of the memory coefficients has a similar exponential shape for all our runs, and the memory effect has completely disappeared by the fifth injection. The scaling to VSMOW was performed using the DEST, HIS and ANTA standard waters of known isotope composition and using a linear regression to correct for potential deviations from the 1:1 line. A fourth intermediate standard water (HERA, working standard 5A) was used for quality control, and we found a precision of ±0.2 (SE of the difference from true value) in δ¹⁸O and ±1.1 in δ²H.

Liq-CRDS and IRMS analysis

Water samples (from sampling (2), bulk water from one whole half of the leaf) were analysed using both IRMS and conventional CRDS (coupling the CRDS analyser with an autosampler and a vaporizer). In this study, we use the mature conventional CRDS technique as a stepping stone to compare the new IM-CRDS technique, and the proved, but expensive IRMS analysis. Indeed, we found that standard waters analysed on IM-CRDS and conventional CRDS showed similar results (data not shown). In addition, users of the IM-CRDS technique are likely to be equipped for conventional CRDS analysis, but many might not have easy access to a mass spectrometer. By comparing all three set-ups, we aim at providing a useful resource for people without access to IRMS.

Ten samples of liquid water extracted from C. esculenta leaf and stem samples were sent to the Center for Stable Isotope Biogeochemistry at the University of California in Berkeley, USA, for IRMS analysis. Deuterium was obtained by chromium combustion using a hot chromium reactor unit (H/Device, similar to the one used in A. G. West et al. (2010), Thermo Finnigan, Bremen, Germany): the water sample was injected in the H/Device and reduced to H₂ gas. The ratio of D/H was then measured on a Thermo Delta Plus XL mass spectrometer with a precision of ±0.60. For the ¹⁸O analysis, water from standards and samples was pipetted into glass vials and quickly sealed. The vials were then purged with 0.2% CO₂ in helium and allowed to equilibrate at room temperature for at least 48 h. The ¹⁸O in the CO₂ taken from the overhead in the sample vials was then analysed by continuous flow using a Thermo Gas Bench II interfaced to a Thermo Delta Plus XL mass spectrometer with a long-term external precision of ±0.10 (Wenbo Yang, private communication). The values of the standards used for the IRMS calibration were (δ²H/δ¹⁸O): SPW3: −251/−31.77‰; CSIB: −88.3/−12.21‰; BWW: −35.5/−4.77‰; and BSMOW: 6.0/4.63‰.

Aliquots of the samples analysed on IRMS were also analysed by conventional CRDS at the Ecohydrology Laboratory at Princeton University, USA. For this analysis, 1.8μL of water was injected into an A0211 high-precision vaporizer by Picarro Inc., and the vapour was pushed with dry air through an MCM similar to that of the IM, before reaching the analyser. Concentrations of O, O and HD¹⁶O were measured on a CRDS analyser (L2103-i) from Picarro Inc. In the following, this method will be referred to as liq-CRDS.

Because we could only analyse a limited number of samples on IRMS, we used a four-step method in order to compare the IM-CRDS and the IRMS techniques. (1) The ten samples analysed by IRMS were also analysed by liq-CRDS, and we calculated the regression between the two techniques (Fig. 2) as described by the following relation:

(2)

Equation 2 was used to correct the cryogenically extracted bulk water analysed on liq-CRDS. (2) To obtain an average leaf value from the many punch holes of the IM-CRDS sampling scheme, we calculated a weighted average of all the samples. The weight to be attributed to each sample was proportional to the area corresponding to each sampling location as determined by a closest-neighbour interpolation (Supporting Information Fig. S3). (3) We compared the values obtained from the extracted water of the half-leaf analysed by IRMS to the bulk leaf water composition obtained on the half-leaf analysed by IM-CRDS (Fig. 3). (4) We included a correction to take into account the difference in water sampled by IRMS and IM-CRDS: the punch holes used in the IM-CRDS method were exclusively sampled within the lamina, whereas cryogenically extracted water used for IRMS analysis included vein water. Vein water from the petiole of all leaves was obtained by cryogenically extraction and analysed by liq-CRDS (δ_vein, Table 1), assuming negligible enrichment in the main veins. To estimate the relative contribution of lamina and vein water, we separated, weighted, oven dried and re-weighted the lamina, the midrib and the secondary veins of eight C. esculenta leaves with lengths ranging from 14 to 64 cm. We determined that the lamina, midrib and the secondary veins represent 62.8 ± 1.2, 11.5 ± 0.4 and 25.7 ± 1.3% of the total leaf water mass, respectively (Fig. 4). Average isotope composition of the leaf water obtained by IM-CRDS (δ_lamina) was corrected using a simple mass balance equation to include the effect of vein water and obtain an estimate of the bulk leaf water composition (δ_bulk). Here, bulk describes whole-leaf water, combining lamina and vein water.

(3)

Isotope mapping

Once sampling (1) was completed, the leaves were pressed and scanned. The position of the midrib, the secondary veins and the sampling locations was hand digitized on the image of the leaf. The data were checked for normality, and a White's (1980) test was performed, and no heteroskedasticity was detected. We interpolated the isotope composition of the leaf in-between sampling locations using an inverse distance weighting (IDW) method with a fixed weight parameter available in the ‘gIDW’ add-on package developed by Giuliano Langella for matlab (Mathworks Inc., Natick, MA, USA). The IDW method was chosen because it is commonly used to produce isoscapes (J. B. West et al. 2009). The interpolation was used to help visualizing the isotope patterns within the leaf, but only the actual samples and their locations were used for comparison with the models, so our results are not dependent on the interpolation technique. In order to obtain a global estimate of the uncertainty associated with the IDW interpolation, we used an n-1 jackknifing method (Bowen & Revenaugh, 2003): the interpolation was rerun n times, each time removing one of the data points and using the remaining n-1 data points for the interpolation. We then compared the estimated value at the sampling location with the actual value and provided the root mean squared error (RMSE) associated with each map.

Because leaves had different vein water isotope compositions (Table 1), we used the enrichment above source water, Δ_l, to compare the different leaf maps:

(4)

where R_sample and R_source are the isotope ratios of the lamina sample and the vein water, respectively, and are defined as follows:

(5)

where R_VSMOW is the heavy-to-light isotope ratio of the international reference standard, here VSMOW, and is equal to 0.00015576 for ²H/¹H and 0.0020052 for ¹⁸O/¹⁶O (Fry, 2007). Within a leaf, Δ_l¹⁸O_max and Δ_l²H_max are defined as follows:

(6)

where Δ_leaf is the array containing all individual Δ_l values within the leaf.

Environmental and physiological data

Environmental data were collected at the Butler Roof Station located about 500 m away from the location of the plants and maintained by the Hydrometeorology Research Group at Princeton University (http://hydrometeorology.princeton.edu/#stations). The station records air temperature, relative humidity, rainfall, wind speed and upward and downward shortwave radiation. Data are available hourly for the entire duration of the experiment (11 July to 30 July 2014). Stomatal conductance and leaf temperature were recorded prior to leaf sampling using an SC-1 Leaf Porometer (Decagon Devices Inc., Pullman, WA, USA). Six random locations around the edge of the lamina (the porometer could not reach far into the leaf) were measured each time, and the average value of all six locations at each sampling time is used here. Environmental and physiological data for each leaf are listed in Table 1.

Modelling leaf isotopes spatial patterns

The SoL model adapted from Gat & Bowser (1991) for an infinite number of evaporative elements in Farquhar & Gan (2003) is defined as

(7)

with Δ_leaf the isotopic enrichment of the water at the evaporative site relative to source water as described by Eqn 4, l the distance from the petiole, l_max the maximal distance from the petiole within the leaf and h′ a parameter defined as h′ = 1+ α⁺α_k(1 − h). Here, h is the relative humidity, α⁺ is the isotope equilibrium fractionation factor (note that α⁺ = 1 + ε⁺) and is calculated using the temperature-dependent expression from Horita & Wesolowski (1994) using temperatures described in Table 1. α_k is the kinetic fractionation factor (again α_k = 1+ ε_k) and is taken to be constant and equal to (Merlivat, 1978) and . Δ_M is the maximum possible enrichment and is defined as

(8)

with Δ_v the isotope composition of atmospheric water vapour, relative to source water (Eqn 4). The isotope composition of the water vapour was obtained by analysing ambient air on the analyser.

The FG model was initially designed for leaves with long and disconnected veins, but it successfully predicted leaf water enrichment in cotton leaves (Ripullone et al. 2008). According to the authors, the aims of the FG model was to take into account the two-pool model represented by veins and mesophyll water, the effects of the backward diffusion of enriched water (Péclet effect) and the progressive enrichment along a line of evaporation. Here, we use the expression for a leaf segment (Gan et al. 2003), which includes the weighted effects of lamina and vein water:

(9)

where ϕ_x is the mass fraction of bulk leaf water coming from the veins. From Fig. 4, we determined that ϕ_x = 37.2 ± 1.2%. The expressions of the ratio of vein water enrichment above source water, Δ_x, and of lamina water enrichment above source water, Δ_la, are as in Gan et al. (2003):

(10)

(11)

where P, P_r and P_l are the lamina radial, the total radial and the longitudinal Péclet numbers, respectively. ₁F₁[a, b; z] is the Kummer function and can be computed using Mathematica (Wolfram Research Inc., Champaign, IL, USA), for example. Here, P, P_r and P_l are fitted to the data of a single leaf at a time using Mathematica.

Testing the induction module

For the five leaves analysed by IM-CRDS for which bulk water was available, we compare the half-leaves analysed by IRMS with the bulk estimates of the other half of the leaves obtained from the IM-CRDS analysis (see Section on Liq-CRDS and IRMS analysis for details). The average difference between IRMS and IM-CRDS was 9.4 ± 1.4‰ in δ¹⁸O (mean ± SE) and 16.4 ± 5.0‰ in δ²H (Fig. 3). This difference is larger than, and therefore not attributable to, the difference observed between the two halves of the same leaf (0.3 ± 0.2‰ in δ¹⁸O and 1.9 ± 1.2‰ in δ²H) in four C. esculenta leaves cryogenically extracted (vein + lamina water) and analysed on liq-CRDS. When correcting the IM-CRDS analysed samples to take into account the influence of the vein water using a simple mass balance (Eqn 3), we obtained that the difference between the IM-CRDS and IRMS techniques dropped to 2.6 ± 0.83‰ in δ¹⁸O and 3.4 ± 2.1‰ in δ²H (Fig. 3).

Spatial patterns

On average, the lamina isotope composition obtained by IM-CRDS varied between 8.6 ± 0.5 and 17.6 ± 0.9‰ in δ¹⁸O and −3.9 ± 1.2 and 16.7 ± 2.0‰ in δ²H. The maps of the isotopic enrichment above source water of six different C. esculenta leaves (Fig. 5, RMSE ( / ): leaf 1: 3.2/22.3‰; leaf 2: 1.85/13.8‰; leaf 3: 2.2/34.9‰; leaf 4: 3.5/4.2‰; leaf 5: 2.7/9.2‰; and leaf 6: 1.4/2.9‰) show generally similar patterns of increased enrichment when moving away from the midrib towards the edge of the leaf. This pattern is observed for both ¹⁸O and ²H. While all leaves cover a similar range of isotopic enrichment from midrib to edge of 5, the leaves show different absolute values, despite being normalized for source water isotope composition, here taken to be equal to petiole water (Table 1). This difference can be explained by the wide range of deuterium excess (d-excess) values observed in the different leaves. D-excess evaluates the deviation from the global meteoric water line and is defined as follows: d-excess = δ²H − 8 δ¹⁸O. D-excess measures the influence of kinetic fractionation of water isotopes and can sometimes provide information on the transpiration regime (Voelker et al. 2014), with more negative values of d-excess indicating more transpiration from the leaf. In our case, leaves with higher isotopic enrichments also show a more negative d-excess (Fig. 5, bottom row), indicating that despite being grown in similar conditions, some leaves experienced more transpiration likely because of small differences in sun exposure during growth. These differences obscure the similitude in relative enrichment throughout the leaf.

Maps of the same leaves with values normalized by the maximum enrichment within each leaf (Supporting Information Fig. S4, RMSE ( ): leaf 1: 28.0/27.9%; leaf 2: 29.3/29.4%; leaf 3: 25.3/32.4%; leaf 4: 42.8/44.6%; leaf 5: 22.6/25.6%; leaf 6: 21.5/23.0%; and combined leaves in Fig. 6: 30.8/31.5%) help remove the dependence to transpiration rate prior to the beginning of the experiment and highlight the general lack of enrichment along the entire length of the midrib. In contrast, enrichment increases rapidly when moving radially, that is, perpendicularly from the midrib. Based on the data presented in Fig. 6, Table 3 summarizes the enrichment at three locations along a radial and a longitudinal gradient in the leaf. A significant enrichment is found along the radial direction with an increase of enrichment above source water by c. 30 percentage points from the midrib to the edges (Table 3), but no pattern is found for enrichment in the longitudinal direction along the axis of the midrib.

Table 3. Percentage from maximum enrichment above source water (%) and (%) for the lamina water at three position along the radial direction (columns) and the longitudinal direction (rows) using the maps shown in Fig. 6

	Midrib			Middle			Edge
	n	Δl18Omax	Δl2Hmax	n	Δl18Omax	Δl2Hmax	n	Δl18Omax	Δl2Hmax
Bottom	13	33.1 (2.0)	30.8 (2.3)	14	49.0 (1.9)	50.3 (2.3)	14	63.4 (2.6)	64.8 (2.5)
Middle	13	26.6 (1.6)	34.7 (1.9)	19	45.7 (1.1)	44.9 (1.4)	14	69.8 (1.5)	68.1 (1.8)
Tip	11	29.6 (2.9)	44.3 (2.2)	10	56.9 (1.7)	55.0 (2.8)	13	58.2 (2.0)	51.6 (2.3)

• Values in parentheses show the standard error. n is the number of data points considered, and the standard error is given in parentheses. In the radial direction, areas represented by each section are c. equal. In the longitudinal direction, areas were split by dividing the total length of the leaf in three, which lead to the ‘tip’ to represent a slightly smaller area because of the tapering of the leaf at its tip. Overall, enrichment in the longitudinal direction is inconsistent, whereas a strong progressive enrichment is found in the radial direction.

Testing leaf water isotope models

Both the SoL and the FG models were tested against our data to determine whether the models could capture the radial enrichment observed in C. esculenta leaves. The SoL model with infinite number of evaporative elements (Farquhar & Gan, 2003) was chosen instead of having a finite number of evaporative elements, as it is most appropriate for leaves in which transpiration occurs in virtually all photosynthetically active tissues. We tested the model against three different l/l_max parameters (example of leaf 5 in Fig. 7): the longitudinal distance l_lon is taken to be the distance from the petiole along the axis represented by the midrib. The radial distance l_rad is the distance from the midrib, measured perpendicularly to it. The total distance is defined as the total distance from the petiole, calculated as . For the longitudinal distance, the SoL model exhibits no significant correlation with the actual data (R² = 0.02, P = 0.5 for student t-test for Fig. 7; see Table 4 for the other leaves). In the case of the radial and the total distance models, the SoL shows a stronger correlation with the data (R² = 0.38, P < 0.001, slope of the best linear regression = 0.27 and R² = 0.57, P < 0.0001, slope of the best linear fit = 0.78, respectively. See Table 4 for the other leaves), but overestimates the enrichment at the edges. The overestimation of the enrichment at the edges by the desert river version of the SoL model has already been pointed out by Gan et al. (2003) and can be corrected for by including Péclet effects into the model. In addition, the lack of lamina–vein interaction in the SoL model means that it cannot capture the variations observed in large reticulate-veined leaves such as C. esculenta, in which the pattern of enrichment is driven by long dense minor veins.

The FG model, by incorporating a much more complex system of leaf water enrichment, is expected to better capture the observed data (example of leaf 5 in Fig. 8). P, P_r and P_l are fitted to the data using Mathematica, and the values obtained for the different Péclet numbers are presented in Table 4. Similarly to the SoL model, there is no significant relation between the longitudinally ordered data and the longitudinal FG model (for the example shown in Fig. 8: R² = 0.09, P = 0.2, slope of the best linear fit = 0.27), with most of the data being largely underestimated by the model. There is however a significant relationship between the data and the model for both the radial and the total distance case, and unlike the SoL model, the slope of the regression for the FG model is always very close to 1 (radial case: R² = 0.46, P = 0.001, slope of the best linear fit = 1.00; total distance case: R² = 0.60, P < 0.0001, slope of the best linear fit = 0.99). The radial model is consistently performing better than the longitudinal model (Table 4), except for leaf 4, which shows no significant pattern in either direction (also visually appreciated in Fig. 5). In leaf 1, the model appears to be strongly influenced by a very depleted sample close to the petiole and a very enriched sample towards the tip. The values of P_l, P and P_r are in agreement with previous values found in maize (Gan et al. 2003). Their relative differences agree with the anticipated patterns, with P_l expected to be the largest of all three Péclet numbers because of the large advection mechanism entraining water in the major veins, while P is expected to be the smallest because it is associated with dominantly diffusive processes in the leaf lamina (Cernusak et al. 2016).

Applications of the IM-CRDS technique

The discrepancy observed between IM-CRDS and IRMS analysis suggests that further testing is still necessary before using the IM-CRDS technique as an absolute method (Fig. 3). In particular, it would be interesting to inject extracted leaf water samples onto filter paper for IM-CRDS analysis and compare these results with IRMS. However, the method is appropriate when using isotopically enriched water as tracers, as well as in situations such as this case study, in which the range of variation observed is about one order of magnitude bigger than the range of variation of the difference between IM-CRDS and IRMS. Some of the observed differences between the two techniques might be due to some evaporation in the sampled punch holes used for the IM-CRDS, which did not occur for the whole-leaf extraction used for the IRMS analysis. Freezing the leaves before sampling might help reducing evaporation, although freezing leaves as large as for C. esculenta might present other logistic challenges. Freeze-clamping the leaf using liquid nitrogen cooled instruments might be another alternative.

Understanding the radial isotopic enrichment in C. esculenta

The lack of longitudinal enrichment and the strong radial enrichment gradient observed in Fig. 5 differs from previous results of leaf water isotopic enrichment in monocots (Helliker & Ehleringer, 2000; Gan et al. 2003; Ogee et al. 2007) that showed that the midrib was the main axis of enrichment. For C. esculenta, the observed pattern is likely due to the large proportion of water contained in the veins, in particular, the midrib (Fig. 4). The enrichment from transpiration does not affect the vein water composition, because of the relatively small losses of water compared with the large water content of the vein, as well as the low hydraulic resistance, and therefore large flux in the midrib (Cochard et al. 2004). Using Eqn 10, we calculated that the expected enrichment along the midrib for the order of magnitude of the Péclet numbers obtained (Table 4) should be less than 1 (Supporting Information Fig. S5). This is in agreement with the results from Table 3 that show that no significant enrichment occurs along the longitudinal direction. Indeed, the isotopic enrichment happens along the axes of the secondary veins, where we see an increase in enrichment above of source water between the midrib and the edge of the leaf (Table 3). Radial isotope enrichment has been observed before in banana leaves (Cernusak et al. 2016) and cotton leaves (Gan et al. 2002). In addition, giant leaves similar to C. esculenta have been shown to have a lower major/minor vein ratio, which in turn leads to a decrease in hydraulic supply when moving from the midrib towards the edges (Li et al. 2013). Leaves with a higher minor vein density have an increased effective path length from vein to evaporative site (Holloway-Phillips et al. 2016), therefore increasing enrichment when moving away from the midrib and towards the edges, as also seen in Cernusak et al. (2016).

The lack of longitudinal enrichment can also be linked to (1) the large amount of water inside the midrib, which likely increases advection and decreases the relative effect of back diffusion of enriched water from the secondary veins into the midrib, and (2) the large pressure drop between major and minor veins (Zwieniecki et al. 2002), which inhibits back diffusion from the minor veins into the secondary veins. The longitudinal enrichment observed in maize and grass leaves is related to the very different vein structure of these leaves compared with C. esculenta. In these leaves, the connection between parallel major veins is minimal. Instead, each parallel vein is directly influenced by the back diffusion of enriched water from the lamina adjacent to it. Moreover, unlike C. esculenta where the midrib holds a large portion of the leaf water, the major veins are frequent and numerous in maize (distance between lateral veins: 1.7 mm) and wheat (distance between lateral veins: 0.8 mm (Gan et al. 2003), compared with several centimetres for C. esculenta), therefore decreasing the relative water held in each vein. This might in turn further increase back diffusion effects, therefore intensifying longitudinal enrichment.

Finally, for giant leaves such as C. esculenta, the differences in boundary layer effects cannot be ignored. In low wind conditions such as the courtyard in which the plants were grown for this study, or the understory they belong to in the wild, the leaf boundary layer is expected to be thicker at the centre of the leaf than at the edges (Schuepp, 1993), leading to an increase in transpiration at the edges, causing increased enrichment.

Testing the ability of the SoL and the FG models to capture the observed radial enrichment, we find that the FG model was adapted to reproduce the pattern, but that this was not true of the SoL model. The SoL model has been successful in capturing spatial patterns in grass (Helliker & Ehleringer, 2000), but its limitations were quickly pointed out (Gan et al. 2002; Farquhar & Gan, 2003). The addition of a two-dimensional Péclet effect in the FG model took into account the progressive enrichment of the xylem (Farquhar & Gan, 2003) and was successfully tested on maize leaves (Gan et al. 2003). With the exception of the SoL tested on cotton leaf (Gan et al. 2002), both models have so far only been tested on monocots with longitudinal veins and high major/minor vein ratios like maize (Gan et al. 2003; Affek et al. 2005; Ogee et al. 2007) and grass (Helliker & Ehleringer, 2000; Ogee et al. 2007). Here, we showed that the models can be applied to more complexly veined leaves, such as C. esculenta. Most importantly, the values of the Péclet numbers found give us an insight into water transport mechanisms through the leaf, with advection dominating transport along the midrib, and slow, diffusion processes happening at the interface between secondary veins and the lamina, as well as within the lamina. The slow diffusion of water through the lamina makes the water more prone to enrichment because of the longer residence time of water close to evaporative sites.

Implications

Transpiration patchiness could also in part explain the patterns in isotope composition observed in the results. Indeed, the leaf edges of such large leaves are subjected to very different temperature and wind speed than the rest of the leaf, leading to changes in stomatal conductance. Under high leaf temperature, stomatal conductance in giant leaves has been found to decrease towards the edges of the leaf (Li et al. 2013), likely due to increased vapour pressure deficit at the edges that would drive water losses. This effect would likely be dependent on leaf size, as leaf temperature and boundary layer vary strongly with leaf size. On the other hand, leaf hydraulic resistance does not correlate with leaf size (Sack & Frole, 2006). We anticipate that under moderate leaf temperature, smaller leaves with similar venation patterns will exhibit comparable leaf water isotope distribution.

The isotope composition of the leaf organic matter is directly linked to leaf water isotopes, because of the exchange of oxygen between water and carbonyl groups (Lehmann et al. 2017). However, comparisons of leaf water and dry matter within one leaf have showed that the isotopic enrichment of organic matter is usually less pronounced than that of leaf water (Barbour, Schurr, et al. 2000; Gan et al. 2003; Kimak et al. 2015), possibly because of the influence of sucrose coming from other leaves and used during the early stages of leaf development. Consequently, the pattern of leaf water enrichment observed here is likely to be attenuated in leaf organic matter.

The photosynthetic capacity of leaves has been found to consistently correlate with vein patterns across many different species (Brodribb et al. 2007; Ocheltree et al. 2012). The photosynthetic capacity is also known to correlate with stomatal conductance (Brodribb et al. 2005), and we would therefore expect stomatal conductance to increase radially when moving away from the midrib. Patchiness in stomatal conductance has been widely observed in a number of different species (Terashima, 1992), but the exact mechanism by which the patchiness develops is yet to be fully understood (Mott & Peak, 2006). Some evidence has been found that even if stomata density does not appear to vary within the leaf, guard cell length significantly decreases when moving from the midrib towards the edge of the leaf (Li et al. 2013), which would lead to a decrease in leaf conductance to water vapour (Nardini et al. 2008) and an increase in leaf temperature at the edges. The heterogeneity of produced sugars due to the differences in stomatal conductance would in turn have consequences for sugar loading strategies (Reidel et al. 2009; Fu et al. 2011), an area of research still at its infancy and where new evidence now points at the possibility of multiple loading strategies coexisting within a single leaf (Slewinski et al. 2013).