Relationship of leaf oxygen and carbon isotopic composition with transpiration efficiency in the C₄ grasses Setaria viridis and Setaria italica

Growth and greenhouse conditions

Gas exchange measurements

Measurements were made on the uppermost fully expanded leaf between 11:00 h and 15:00 h. Leaves were placed in a 2 × 3 chamber of an LI-6400XT open gas exchange system (Li-COR Biosciences, Inc., Lincoln, NE, USA). The leaf was allowed to acclimate at 1500 µmol m⁻² s⁻¹ PPFD, leaf temperature of 29 °C, flow rate of 300 µmol s⁻¹, 21% O₂, 35 Pa CO₂ for S. viridis. The same conditions were used for S. italica except that light intensity was 900 µmol m⁻² s⁻¹ PPFD to reflect the light intensity of their growing conditions. Relative humidity (RH) in the LI-COR chamber was within 10% of the RH under growth conditions, and the implications of this difference in RH is explained in the next section.

Sample collection for stable isotope analysis

To obtain sufficient leaf water to measure δ¹⁸O of leaf water (δ¹⁸O_LW) from S. viridis, an aggregate of five to eight leaves (including the leaf used to measure gas exchange) was collected from each plant at the time of harvest. However, in S. italica the leaf used for gas exchange measurements was sufficient to analyse δ¹⁸O_LW. The same leaf samples that were analysed for δ¹⁸O_LW were also analysed for leaf carbon isotopic composition (δ¹³C_leaf) and oxygen isotopic composition of bulk leaf tissue (δ¹⁸O_BL). All leaves collected for stable isotope analysis were the youngest, fully expanded leaves, which developed 15–20 d after soil water content reached the treatment set point. For both species, leaves were removed from the plant and photographed to measure leaf area using ImageJ software (Schneider et al., 2012). Photographing each leaf only took approximately 20 s, and then the leaf was stored in sealed glass tubes awaiting water extraction.

In S. italica only, using the same leaf for gas exchange measurements and stable isotope analysis could have an effect on δ¹⁸O_LW because RH and the oxygen isotope ratio in the growth cabinet air (δ¹⁸O_v) could differ between growth and gas exchange chambers (Loucos et al., 2015). However, gas exchange measurements were conducted within the growth chamber, and the mean±SE proportion of the total leaf area continuously exposed to growth cabinet conditions during gas exchange measurements was 94.0 ± 0.6%, 91.0 ± 0.8%, and 89.5 ± 0.8% for well-watered, moderately water-limited, and severely water-limited treatments, respectively. Therefore, the δ¹⁸O of the growth cabinet air was the most appropriate measure of δ¹⁸O_v. The difference in RH between the growth cabinet and the gas exchange chamber could contribute a 1.9–5.5‰ shift in δ¹⁸O_e for the leaf section in the gas exchange chamber, assuming, however unlikely, that 10–20 min was adequate time for the leaf section to acclimate. This would represent a shift in δ¹⁸O_e for the entire leaf of 0.24–0.43‰, and Δ¹⁸O_LW would shift by a fraction of this. Nonetheless, taking into account this potential shift did not significantly influence observed Δ¹⁸O_LW values across treatments. Differences between greenhouse and LI-COR chamber conditions for S. viridis were irrelevant because leaf water was extracted from an aggregate of five to eight leaves.

The δ¹⁸O of water vapor in the greenhouse and growth chambers was measured every 30 min during gas exchange measurements by collecting air in 5-liter Supel inert foil gas sampling bags (Supelco, Bellefonte, PA, USA). Bags were flushed several times with air before filling, and δ¹⁸O of water vapor was immediately measured on the cavity ringdown spectrometer (L1102-i water analyser, Picarro Inc, Santa Clara, CA, USA).

Following the gas exchange measurements, root crowns were cleaned of soil and stored at −20 °C in air-tight tubes. Root crowns are considered the bottom 1 cm of the culm or tiller, but no actual roots were collected. Additionally, two soil samples were collected at the top and bottom of each pot, and stored using the same method as with the root crowns. Leaves, soils, and root crowns were distilled using a cryogenic vacuum distillation method (Vendramini and Sternberg, 2007). Additionally, daily samples of irrigation water (IW) were collected to measure δ¹⁸O_IW.

Biomass measurements

After collection of plant tissue for stable isotope analysis, the entire aboveground biomass was collected and weighed for fresh weight. Samples were dried at 65 °C for 3 d before weighing dry biomass.

Stable isotope analysis

The stable isotope composition of carbon and oxygen (δ¹⁸O and δ¹³C, respectively) were reported in δ notation in parts per thousand (‰),

where R_sample and R_standard are the molar ratios of heavy to light isotope (¹⁸O/¹⁶O and ¹³C/¹²C) of the sample and international standard, respectively. The international standard used for oxygen was Vienna Standard Mean Ocean Water (VSMOW) and for carbon was Vienna Pee Dee Belemnite (VPDB).

Stable isotope analysis by isotope ratio mass spectrometer

Leaf tissue was analysed for oxygen isotopic analysis by converting to CO with a pyrolysis elemental analyser (TC/EA, Thermo Finnigan, Bremen, Germany) and analysed with a continuous flow isotope ratio mass spectrometer (Delta PlusXP, Thermo Finnigan; Brenna et al., 1997^; Gehre and Strauch, 2003). For carbon isotopic analysis, leaf tissue was converted to CO₂ with an elemental analyser (ECS 4010, Costech Analytical, Valencia, CA, USA) and analysed with a continuous flow isotope ratio mass spectrometer (Delta PlusXP; Brenna et al., 1997^; Qi et al., 2003). Isotopic standards, Standard Light Antarctica Precipitation (SLAP; Gonfiantini, 1978) and Puerto Rico water (Qi et al., 2014), were analysed alongside samples to calculate δ¹⁸O based on the VSMOW scale. Lab standards, calibrated to international standards, were used to calculate δ¹³C relative to VPDB. Standard error for δ¹³C values for the S. viridis and S. italica experiments was 0.11 and 0.05‰, respectively. The standard error for δ¹⁸O was 0.2‰ for both experiments.

Leaf and root crown water were analysed for oxygen isotope composition by equilibrating 0.5 ml of water at room temperature with 0.3% CO₂:He mixture for 48 h on a ThermoFinnigan GasBench II (Thermo Electron Corp., Bremen, Germany). CO₂ was analysed with a continuous flow isotope ratio mass spectrometer (Delta PlusXP; Brenna et al., 1997^; Qi et al., 2003).

Stable isotope analysis by isotope ratio infrared spectroscopy

Soil and irrigation water were measured by isotope ratio infrared spectroscopy (model L1102-i, Picarro, Sunnyvale, CA, USA) connected to a vaporization chamber (V1102-i). The mean δ¹⁸O was calculated of the last three of six consecutive analyses on each sample. Three laboratory standards, calibrated to the VSMOW scale, were interspersed among the samples and were used to correct the sample δ¹⁸O values to the VSMOW scale. Water vapor was analysed for at least 15 min, but the mean of the last 5 min was used for δ¹⁸O and corrected to the VSMOW scale.

Calculations of transpiration efficiency

TE_{instantaneous} (A_net/E) and TE_intrinsic (A_net/g_s) are derived from gas exchange measurements, but they are independent of both TE_plant and TE_w, which are independent of each other. TE_plant was derived from whole-plant measures of biomass and transpiration. TE_w was calculated using C_i/C_a derived from δ¹³C_BL, and the calculations are independent of the gas exchange measurements (described below).

Discrimination (Δ¹³C_BL) and C_i/C_a are related in Eq. 2 and were used to calculate the integrated C_i/C_a over the life of the leaf (Henderson et al., 1992^, 1998). A constant leakiness (ϕ) of 0.21 was assumed for all plants. The δ¹³C of ambient CO₂ in the greenhouse and growth chamber that was used to calculate Δ¹³C_BL was −10.7 ± 0.8‰, which was collected various times over a period of several months that included the time that the experiment was conducted. Air samples were collected in 5-liter Tedlar gas sampling bags using the same collection procedure used to collect air vapor. The gas was analysed by introducing air directly into either the isotope ratio mass spectrometer or the tunable diode laser absorption spectroscope (model TGA 200A, Campbell Scientific, Inc., Logan, UT, USA; Ubierna et al., 2013). Transpiration efficiency (TE_w), defined as the ratio of dry matter produced per unit of water transpired, was calculated from the δ¹³C_BL as described in Henderson et al. (1998) as:

where v is the leaf to air vapor pressure difference and ϕ_r was calculated as the measured ratio of respiration and photosynthetic rate (0.08 for all plants). This parameter was calculated from gas exchange measurements made during the experiment. The parameter ϕ_w was calculated as the measured ratio of whole-plant night to day transpiration (0.30 and 0.23 for water-limited and well-watered plants, respectively) measured in this experiment. Both night-time and daytime whole-plant transpiration were measured on S. viridis. Both ϕ_r and ϕ_w were measured for S. viridis and assumed to be the same for S. italica. C_i/C_a was calculated from δ¹³C_BL using Eq. 2.

Model calculations to determine validity of the Péclet model

To test the applicability of the Péclet model and its relationship to E, we used the method described by Holloway-Phillips et al. (2016) and Song et al. (2015) of determining the proportional deviation (f) of Δ¹⁸O_LW from Δ¹⁸O_e plotted against E where f is calculated as:

Effective path length (L) was calculated using the equations described in Song et al. (2013).

Statistical analysis

In both experiments statistical analyses were conducted in R v. 3.3.0 (R Core Team, 2013), using car (v. 2.0-26) and agricolae (version 1.2-2) packages for statistical tests. Model II regressions (standard major axis regression) were calculated, using the lmodel2 (v. 1.7-2) package, because neither variable was controlled, both varied naturally with their own associated error, and the physical units of both variables were not the same. Homogeneity was tested based on plotting predicted fit vs residuals. Using the extRemes package (v. 2.0-8), normality was tested by plotting residuals on quantiles–quantiles plots. In all cases, where normality was questionable, transforming the data did not change the statistical results, so the data were not transformed. One- and two-way analysis of variance (ANOVA) was used to determine differences across treatments in the experiment with S. italica and between treatments and collection periods for S. viridis. Two-sample Student’s t-tests were performed to determine the difference between δ¹⁸O_SW and δ¹⁸O_RC values. One-sample t-tests were performed to determine if the difference of both δ¹⁸O_SW and δ¹⁸O_RC with δ¹⁸O_IW was significantly different from 0. Repeated measures ANOVAs were conducted on the following parameters: total plant water use and GWC for S. viridis and on daily water use, GWC, A_net, E, and g_s for S. italica.

Plant growth and treatment effect

Maintaining water-limited plants at a GWC 71% lower than the well-watered plants significantly reduced total water use by 59% (Table 2 and Fig. 1). Additionally, at all collection times fresh and dry aboveground biomass were lower in the water-limited treatment relative to the well-watered treatment (Table 3). The results for S. italica were similar to S. viridis in that the GWC was reduced by 72% and 82%, and total water use by 70% and 80% in the moderately and severely water-limited treatments, respectively (Table 2; Fig. 1). In S. italica, fresh and dry aboveground biomass were reduced in the moderately water-limited, 71% and 68%, respectively, and in the severely water-limited treatment, 79% and 74%, respectively (Table 4; Fig. 1). Additionally, leaf length in the water-limited treatments was shorter compared with the well-watered leaves by 18% in S. viridis and in S. italica by 32% and 57% in the moderately and severely water-limited treatments, respectively (Tables 3 and 4). The number of tillers per plant in the well-watered treatment was 44% greater than the water-limited treatment in S. viridis (Table 3); however, the number of tillers per S. italica plant did not differ across treatments (Table 4).

Open multimedia modal

Open multimedia modal

Open multimedia modal

Open multimedia modal

For both species, stomatal conductance (g_s), rates of transpiration (E), and the net rate of CO₂ assimilation (A_net) were generally higher in the well-watered treatment compared with the water-limited treatment. Gas exchange measurements of S. viridis were made during the three biomass collections, and E was 33% and 67% greater in the well-watered treatment in collections 2 and 3, respectively, but did not differ between treatments in collection 1 (Table 3). Additionally, g_s was 41% and 76% greater in the well-watered treatment during collection 2 and 3, respectively, but did not differ between treatments in collection 1. However, A_net in S. viridis was different between treatments only in collection 3 when A_net of water-limited plants was 61% lower than that of the well-watered treatment (Table 3). In S. italica, E, g_s, and A_net were not different between the severely and moderately water-limited treatments, but both treatments were on average 27%, 32%, and 16% lower, respectively, than the well-watered treatment (Table 4).

Leaf carbon isotopic composition

The response in leaf carbon isotopic signature (δ¹³C) to water limitations was similar for both species. For example, δ¹³C values in S. viridis were consistently lower in the water-limited treatment for all collections by 1.1‰ (Table 3) and in S. italica the δ¹³C values were 0.9‰ lower in the two water-limited treatments compared with the well-watered treatments (Table 4). The δ¹³C values of both species were positively correlated with A_net, E, and g_s, leaf water content, Δ¹⁸O_LW, Δ¹⁸O_BL, and all measurements of TE and plant growth and had stronger correlations with these parameters than either Δ¹⁸O_LW or Δ¹⁸O_BL (Table 5 and Figs 2 and 3).

Open multimedia modal

Open multimedia modal

Open multimedia modal

Using the simplified equation (Eq. 2) of Δ¹³C versus C_i/C_a, where Δ¹³C was calculated from δ¹³C_BL and C_i/C_a was measured, the mean leakiness of 0.21 ± 0.02 and 0.19 ± 0.01 was calculated for S. viridis in the well-watered and water-limited treatments, respectively. In S. italica leakiness was 0.17 ± 0.01 for well-watered and severely water-limited treatments and 0.20 ± 0.01 in the moderately water-limited. Overall, leakiness did not significantly differ between species or across treatments (0.17–0.21). The difference in leakiness would account for 0.22 ± 0.03‰ and 0.24 ± 0.01‰ of the observed difference in Δ¹³C_BL between treatments over the observed range of C_i/C_a in S. italica and S. viridis, respectively.

Transpiration efficiency (TE) and leaf δ¹³C-derived transpiration efficiency (TE_w)

Four different methods were used to calculate transpiration efficiency: (i) long term TE (TE_plant; grams of aboveground dry biomass per liter water transpired); (ii) instantaneous TE (TE_{instantaneous}; A_net/E); (iii) intrinsic TE (TE_intrinsic; A_net/g_s); and (iv) δ¹³C_BL-derived TE (TE_w; mmoles carbon fixed per mole H₂O transpired). For S. viridis the water-limited plants had higher TE regardless of how it was estimated, except in collection 3 where there was no difference in TE_{instantaneous} between treatments (Table 3). In S. italica, all four estimates of TE were higher in both water-limited treatments than the well-watered treatment, but the difference was not significant between the moderately and severely water-limited treatments. In both species, the TE_intrinsic had the largest differences between treatments (45% and 32% greater in S. viridis and S. italica, respectively; Tables 3 and 4). Additionally, TE_intrinsic had the strongest relationship with Δ¹⁸O_LW, Δ¹⁸O_BL, and δ¹³C (Table 5).

Leaf oxygen isotopic composition

The gas exchange measurements and leaf samples were collected between 11:00 and 15:00 h. During this time the vapor oxygen isotope ratios (δ¹⁸O_v) in both the greenhouse for S. viridis and in the growth chambers for S. italica were relatively stable. For S. viridis, δ¹⁸O_v values (mean±SE) were −17.4 ± 0.2, −21.9 ± 0.4, and −24.2 ± 0.3 for the three collections, respectively. For S. italica, δ¹⁸O_v values were −25.8 ± 0.1 and −20.1 ± 0.8 for the two collections, respectively.

Plants were top irrigated in covered pots with irrigation water (−17.0 ± 0.1‰ and −16.9 ± 0.1‰ for S. viridis and S. italica, respectively). The mean δ¹⁸O_SW values (average of δ¹⁸O of soil samples from the top and bottom of the pot) of the well-watered and water-limited treatments were 0.8 ± 0.1‰ and 1.9 ± 0.2‰ higher than irrigation water for S. viridis (P<0.0001), respectively. For S. italica, δ¹⁸O_SW values were 0.8 ± 0.2‰, 2.7 ± 0.2‰, and 3.9 ± 0.4‰ higher than irrigation water in the well-watered, moderately and severely water-limited treatments, respectively (P<0.0001). In S. viridis, the δ¹⁸O_RC values were 0.6 ± 0.1‰ and 2.0 ± 0.1‰ higher than δ¹⁸O_IR in the well-watered and water-limited plants (P=0.0002 and P<0.0001), respectively. In S. italica, δ¹⁸O values were 1.1 ± 0.1‰, 3.1 ± 0.2‰, and 4.3 ± 0.3‰ higher than the irrigation water in the well-watered, moderately and severely water-limited treatments (P<0.0001), respectively. The δ¹⁸O of root crown and soil water was not significantly different in either S. viridis or S. italica (P>0.05).

Leaf water enrichment (Δ¹⁸O_LW) showed a significant treatment effect in S. viridis independent of which water was considered source water (irrigation, soil, or root crown water), but the treatment effect on Δ¹⁸O_LW was greatest with irrigation water. Likewise the strength of the correlation between parameters of gas exchange, water use, and growth depended on which source water was used to calculate Δ¹⁸O_LW. However, independent of the source water, Δ¹⁸O_LW negatively correlated with A_net, E, and g_s and positively correlated with TE_intrinsic and δ¹³C_BL (Table 5 and Fig. 4). In S. italica, Δ¹⁸O_LW did not have a significant treatment effect or correlate with growth, gas exchange, or TE variables, regardless of source water used (Table 4).

Open multimedia modal

In S. viridis, a significant treatment effect in Δ¹⁸O_BL was only found and Δ¹⁸O_BL only correlated with g_s, E, fresh aboveground biomass, TE_plant, and TE_intrinsic, Δ¹⁸O_LW, and δ¹³C_BL when irrigation water was used as source water (Table 5 and Fig. 4). The only parameter that correlated significantly with Δ¹⁸O_BL when root crown water was considered the source water was Δ¹⁸O_LW. For S. italica, the Δ¹⁸O_BL in the well-watered treatment was 2.0‰ greater than the severely water-limited treatment, but Δ¹⁸O_BL of the moderately water-limited treatment was not different from either treatment (Table 4). This difference resulted in significant positive correlations with E and g_s (0.60 and 0.59, respectively) when negative correlations were expected.

The Péclet model was tested by comparing the proportional deviation of Δ¹⁸O_LW from Δ¹⁸O_e (f; Eq. 6) with E (see Supplementary Fig. S1 at JXB online). This relationship was significant for S. viridis (f=0.086E–0.221, R²=0.36, P=0.0005) but not for S. italica (P=0.93) suggesting that in S. viridis Δ¹⁸O_LW deviated more from Δ¹⁸O_e as E increased. For S. italica, the Δ¹⁸O_LW was larger than Δ¹⁸O_e (negative f values), causing Δ¹⁸O_LW to be more enriched than would be expected based on the evaporative environment. For S. viridis the estimated effective leaf length (L) was small but significantly different between treatments (13.2 ± 2.7 and 8.2 ± 2.4 mm in well-watered and water-limited treatments, respectively; Supplementary Tables S1 and S2).

Leaf carbon isotopic composition

Water-limited C₄ plants consistently have lower δ¹³C_BL values than well-watered plants (Henderson et al., 1998^; Ghannoum et al., 2002). Water limitations also typically reduce g_s, and in C₄ plants low g_s results in depleted δ¹³C_BL. This is primarily because Δ¹³C_BL decreases with C_i/C_a when ϕ is generally below 0.37, and low g_s tends to decrease C_i/C_a (Cernusak et al., 2013^; von Caemmerer et al., 2014). Alternatively, a decreased photosynthetic capacity in the water-limited plants could increase C_i/C_a, which would decrease Δ¹³C_BL and lead to an increase in δ¹³C_BL values. Therefore, the decrease in δ¹³C_BL observed in the water-limited plants is mostly due to changes in g_s and its influence on C_i/C_a. However, it is possible that water limitations increase ϕ, and lower δ¹³C_BL values.

Models of C₄ isotope exchange suggest that a decrease in the capacity of the CO₂ concentrating mechanism, as would occur with a stomatal limitation in CO₂ supply, could decrease ϕ (Ellsworth and Cousins, 2016). Additionally, leaf level measurements of CO₂ exchange have demonstrated that ϕ remains fairly constant under various environmental conditions (Henderson et al., 1998^; Ubierna et al., 2011^, 2013^; Sun et al., 2012^; Cernusak et al., 2013^; Kromdijk et al., 2014^; von Caemmerer et al., 2014). The ability of C₄ plants to maintain and minimize ϕ in response to long-term changes in growth conditions is not surprising as the C₄ and C₃ cycles are metabolically coordinated between the mesophyll and bundle sheath cells, causing them to function as integrated and not independent cycles (Furbank et al., 2013).

Using the simplified model of Δ¹³C (Eq. 2), the calculated ϕ from δ¹³C_BL and C_i/C_a produced similar values (0.17–0.21) to what has been published previously (see list of studies in Kromdijk et al., 2014). This difference in ϕ could account for a mean 22% and 24% of the measured difference in δ¹³C_BL across treatments in S. italica and S. viridis, respectively. In a separate study, under well-watered conditions Δ¹³C_{instantaneous} at similar C_i/C_a was approximately 4.4 ± 0.2‰ for both S. viridis and S. italica, giving evidence that ϕ is not inherently different between these species (Ellsworth et al., unpublished data). Additionally, the relationship between Δ¹³C_BL and C_i/C_a was similar for both species under all treatments, potentially falling on the same line where ϕ controls the slope (see Fig 1; Ellsworth and Cousins, 2016). Granted these calculations of ϕ provide only an approximation because instantaneous measures of Δ¹³C and δ¹³C_BL are known to differ because of post-photosynthetic fractionations, as discussed below (von Caemmerer et al., 2014^; Ellsworth and Cousins, 2016).

Post-photosynthetic fractionation

Post-photosynthetic fractionations of carbon compounds could influence δ¹³C_BL; however, to influence δ¹³C_BL there must be a change in the leaf carbon mass balance by isotopic flux into or out of the leaf (von Caemmerer et al., 2014). Potentially, water-limited plants differ in which carbon pools are exported from the leaf. For example, if enriched amino acids are transported from the leaf at a greater rate than in well-watered plants (Melzer and O’Leary, 1987), then δ¹³C_BL could decrease. However, enriched amino acids are a small pool of carbon compared with sucrose and cellulose, so their export would have to be extremely large to account for the observed depletion in δ¹³C_BL. Additionally, the export or consumption by respiration of sucrose from starch degradation instead of from triose phosphate synthesis could also affect δ¹³C_BL because sucrose from starch degradation is more enriched in ¹³C (Hobbie and Werner, 2004^; Tcherkez and Farquhar, 2005^; von Caemmerer et al., 2014). However, the respiratory carbon flux out of the leaf is relatively small compared with photosynthetic flux into the leaf, so water-limited plants would need an unrealistic shift in the isotopic signature of enriched respiratory carbon source (~12‰) to account for the −1‰ shift observed in δ¹³C_BL. In water-limited C₃ bean plants, sucrose was the primary carbon pool for respiration, suggesting that a dramatic shift in Setaria is unlikely (Duranceau et al., 1999^; Tcherkez et al., 2003). Therefore, as discussed above, the shift in δ¹³C_BL is most likely not due to post-photosynthetic fractionations but rather to the treatment effect on leaf gas exchange and TE.

Leaf carbon isotopic composition across drought experiments

In both species, δ¹³C_BL in water-limited plants consistently showed lower values by 0.9–1.1‰ than in the well-watered plants. Previous studies also have found a difference in δ¹³C_BL values between well-watered and water-limited plants of 0.2–0.6‰, which probably depended on the type, severity, or duration of the reduction in water availability (Saliendra et al., 1996^; Henderson et al., 1998^; Brück et al., 2000^; Monneveux et al., 2007^; Cabrera-Bosquet et al., 2009a). Nonetheless, the consistent depletion in ¹³C in response to water limitations persisted across C₄ species and experiments, lending further evidence that decreased g_s is driving the response in δ¹³C_BL. Additionally, Gresset et al. (2014) found that δ¹³C_BL in C₄ maize was under genetic control, giving support to the potential use of δ¹³C_BL as a genetic screen for TE. However, further research is needed to determine the degree to which δ¹³C_BL can be used to detect subtle differences in TE in C₄ plants, if δ¹³C_BL is under similar genetic control as TE, and if it can be used to screen for TE across genotypes.

Transpiration efficiency estimated from leaf carbon isotopic composition

Transpiration efficiency (TE_w) calculated from δ¹³C_BL correlated strongly with leaf-level gas exchange measurements of g_s and TE_intrinsic (A_net/g_s). Therefore, calculating TE_w based on C_i/C_a estimated from δ¹³C_BL accurately reflected differences in g_s between treatments. However, in S. viridis, TE_plant was more highly correlated with TE_intrinsic and TE_{instantaneous} than in S. italica. This may be because the short and bushy S. viridis has proportionally more leaf biomass than the upright S. italica, so leaf characteristics would have a greater influence on plant-level estimates of TE (e.g. TE_plant) in S. viridis than in S. italica. Nonetheless, for both species, δ¹³C_BL reflected differences in both whole-plant and leaf-level estimates of TE.

Leaf water enrichment

In Setaria viridis, Δ¹⁸O_LW formed a negative relationship with E as expected based on the Péclet model. In the Péclet model, the Péclet number is proportional to E and L, so a positive relationship between Δ¹⁸O_LW and E requires L to remain constant across individuals or treatments. L differed only slightly between treatments, and this did not remove the relationship between Δ¹⁸O_LW and f (1−Δ¹⁸O_LW/Δ¹⁸O_e) with E, showing evidence that the Péclet model best describes leaf water isotopic composition (Supplementary Fig. S1 and Fig. 4). The expected relationship existed for S. viridis, lending strength to the possible use of leaf oxygen isotopic composition as a proxy of E.

Contrary to S. viridis, Δ¹⁸O_LW and E did not form a significant relationship in S. italica, and Δ¹⁸O_LW values were more enriched than Δ¹⁸O_e. One potential reason for this disparity in results between the two species is that leaf temperature differed between treatments and across the range of leaf temperature. High transpiration rate can change Δ¹⁸O_LW and subsequently Δ¹⁸O_BL by decreasing leaf temperature through evaporative cooling. Both ε⁺ and e_i (and therefore e_a/e_i) are temperature-dependent (Barbour, 2007^; Barbour et al., 2004). In the experiment with S. italica, the maximum difference in leaf temperature across the treatments was less than 2 °C (Ellsworth et al., 2016); Δ¹⁸O_e of leaves of water-limited plants would only increase by ~0.6‰, and Δ¹⁸O_LW, being partly composed of Δ¹⁸O_e, would change by a fraction of 0.6‰. If leaf temperature in S. viridis differed less than 2 °C, as previously observed with S. italica (Ellsworth et al., 2016), then the temperature difference between treatments would be insufficient to explain the difference in Δ¹⁸O_LW observed in S. viridis. As for S. italica, Δ¹⁸O_LW values showed the opposite trend as would be expected if the differences were based on leaf temperature. Therefore, the relationship between Δ¹⁸O_LW and E in S. viridis suggests a Péclet effect in S. viridis but not in S. italica.

A possible reason why E does not affect the mixing of source water with water from the sites of evaporation in S. italica (as described in the Péclet model and observed in S. viridis) may be because Δ¹⁸O_LW increases with leaf length in C₄ grasses (Helliker and Ehleringer, 2000). In S. italica, longer leaf length in well-watered plants than in water-limited plants apparently increased Δ¹⁸O_LW values sufficiently to mask the expected relationship between Δ¹⁸O_LW and E. This effect of leaf length was strong enough that Δ¹⁸O_BL was positively correlated with E and g_s. Enrichment up the leaf blade, described as the longitudinal Péclet effect, occurs because the xylem water being supplied to the sites of evaporation becomes progressively more enriched from the base to the tip of the leaf blade (Gan et al., 2003). Therefore, relative to source water entering the leaf base, the water at the sites of evaporation is enriched above what can be attributed to E. In contrast, the treatment difference in leaf length in S. viridis was minimal and insufficient to mask the transpiration-derived differences in Δ¹⁸O_LW.

Bulk leaf enrichment and E

According to theory, bulk leaf enrichment (Δ¹⁸O_BL) should reflect the isotopic signature of leaf water in which bulk tissue is synthesized (Farquhar and Lloyd, 1993^; Sternberg et al., 1986). As expected, a weak but significant positive relationship between Δ¹⁸O_LW and Δ¹⁸O_BL was found in S. viridis, but this relationship did not translate into a significant difference in Δ¹⁸O_BL between treatments or significant correlations of Δ¹⁸O_BL with measures of gas exchange or growth. Three possibilities exist that may explain this pattern. First, the relationship between Δ¹⁸O_LW and E primarily exists because of the longitudinal or xylem Péclet effect and not the mesophyll/lamina or radial Péclet effect. Therefore, the oxygen isotope signature of lamina water that is passed onto organic molecules may have little E-related enrichment, so the relationship between Δ¹⁸O_LW and E would not be passed onto the bulk leaf tissue (Holloway‐Phillips et al., 2016). Second, Δ¹⁸O_LW was measured once, and a single measurement may not capture all variation in E and climatic conditions such as δ¹⁸O_v, relative humidity, and leaf temperature over the leaf lifespan, which can be difficult to control or account for precisely even in a controlled environment setting (Roden and Siegwolf, 2012^; Roden and Farquhar, 2012). Third, leaf length would not affect Δ¹⁸O_BL as much as Δ¹⁸O_LW because Δ¹⁸O_BL would be driven principally by Δ¹⁸O_LW early in leaf construction when oxygen isotope exchange between water and sucrose takes place. Nevertheless, the magnitude of this non-significant difference between treatments was similar to what has been reported in other studies (Cabrera‐Bosquet et al., 2009c^; Sánchez-Bragado et al., 2016). The oxygen isotopic composition of other plant organs has been proposed as proxies because they produced stronger correlations with grain yield than δ¹⁸O_BL (Sánchez-Bragado et al., 2016). However, it is necessary to understand how the leaf oxygen isotope enrichment that is related to E is passed onto these organs before their δ¹⁸O can be used an effective proxy for E.

Another problem that can obscure the relationship that Δ¹⁸O_LW and Δ¹⁸O_BL have with E is misidentifying source water (δ¹⁸O_S) used to calculate Δ¹⁸O_LW and Δ¹⁸O_BL. In this study, we measured the isotopic signature of three possible source waters: (i) irrigation water, (ii) mean soil water, and (iii) root crown water. Isotopically soil and root crown water were statistically indistinguishable, confirming previous studies that there is little fractionation of oxygen isotopes upon uptake by roots, so root crown water is a good representation of δ¹⁸O_S at the time of water collection (Ellsworth and Williams, 2007). Irrigation water does not reflect source water in water limitation studies because it undergoes evaporative enrichment, creating isotopically distinct soil water pools for each treatment. As a result, differences in Δ¹⁸O_LW and Δ¹⁸O_BL between treatments could simply be an artefact of incorrectly identifying δ¹⁸O_S, and not because of other physiological traits or environmental factors. Therefore, care must be taken to define the real source water of leaves.

Acknowledgements

This work was supported by the Office of Biological and Environmental Research in the Department of Energy Office of Science (DE-SC0008769).