2.1 Plant material and growth conditions

Perennial ryegrass (L. perenne, cv. ‘Acento’) plants were grown in four plant growth chambers (PGR15; Conviron) in a 16: 8 h day: night cycle (temperature 20/16°C), under a 3 × 2 factorial design: three atmospheric CO₂ concentration levels (‘half-ambient’ = 200, ‘ambient' = 400 or ‘double-ambient’ = 800 μmol mol⁻¹) and two daytime RH levels (low RH = 50%, high RH = 75%; nighttime RH was 75% for all treatments), as previously described in Baca Cabrera et al. (2020). In brief, L. perenne plants were grown individually in plastic tubes (350 mm height, 50 mm diameter) filled with washed quartz sand (0.3–0.8 mm grain size) and arranged in plastic containers (770 × 560 × 300 mm) at a density of 383 plants m⁻². Plants were supplied four times a day with a Hoagland-type nutrient solution with reduced nitrate-N content. Light was supplied by cool-white fluorescent tubes and warm-white light-emitting diode (LED) bulbs with a constant photosynthetic photon flux density (PPFD) of 800 µmol m⁻² s⁻¹ at plant height during the 16 h-long light period. A total of five sequential experimental runs were performed, resulting in five chamber scale replicates for the so-called ‘reference treatment’ (400 μmol mol^-1 CO₂/50% RH) and three replicate mesocosm-scale runs for the other treatments.

CO₂ and RH treatments were installed on the 13th day after seed imbibition. For this, the air supplied to the chambers was mixed from dry CO₂-free air and ¹⁸O-enriched or ¹⁸O-depleted tank CO₂, using mass flow controllers. ¹⁸O-enriched CO₂ was obtained from CARBO Kohlensäurewerke and had a δ¹⁸O_CO2 which ranged between 14.1‰ and 20.2‰ between individual cylinders; ¹⁸O-depleted CO₂ was purchased from Linde AG and had a δ¹⁸O_CO2 which ranged between −9.0‰ and −2.0‰ between individual cylinders. In each experimental run, treatments were run in duplicate, with one chamber receiving the ¹⁸O-depleted and the other chamber the ¹⁸O-enriched CO₂. RH and temperature were controlled by the chamber control system (CMP6050; Conviron). CO₂ concentration and RH were measured every 30 min by an infrared gas analyzer (IRGA; Li-840; Li-Cor) and never deviated more than ±5 μmol mol⁻¹ and ±2.0% relative to the set nominal value, respectively. The δ¹⁸O of CO₂ at the inlet and outlet of each growth chamber was measured as in Liu et al. (2016) with a continuous-flow isotope ratio mass spectrometer (IRMS) (Delta Plus, Finnigan MAT).

2.2 Sampling design and extraction of tissue water and sucrose

Plants from each chamber-scale replicate were sampled when plant canopies were closed (leaf area index >5, at 7–9 weeks after the beginning of the experiment). At that time, the morphometric traits of leaves were similar among treatments (Baca Cabrera et al., 2020). Sampling took place at c. 2 h before the end of the light and dark periods. In each replicate sampling, 12 plants were randomly selected, dissected and the sampled plant material of 6 plants pooled in 1 subsample (providing 2 subsamples per chamber and per sampling occasion). Additionally, we sampled other plants as described in Baca Cabrera et al. (2020), to estimate LWC (in mol m⁻² leaf area) of young fully expanded leaf blades. LWC was obtained gravimetrically and normalized by the leaf area. The IMAGE J software (Schneider et al., 2012) was used for digital analysis.

For tissue water extraction, the two youngest fully expanded leaf blades and the leaf growth-and-differentiation zone (LGDZ, see Figure 1 in Baca Cabrera et al., 2020) of three mature tillers per plant were excised, sealed in 12 mL Exetainer vials (Labco), capped, wrapped with Parafilm and stored at –18°C until water extraction. Tissue water was extracted for 2 h using cryogenic vacuum distillation as in Liu et al. (2016).

For sucrose extraction, the youngest fully expanded leaf blades of another two mature tillers from the same plants were excised, placed into paper bags, frozen in liquid nitrogen, stored at −18°C until freeze-drying, milled and stored again at −18°C until sucrose extraction. Water-soluble carbohydrates were extracted from 50 mg aliquots of dry material and sucrose separated from other compounds using a preparative high-performance liquid chromatography (HPLC) technique according to Gebbing and Schnyder (2001).

Using separate samples for tissue water and sucrose extraction could theoretically lead to greater scatter in the relationship between δ¹⁸O of leaf water (δ¹⁸O_LW) and sucrose (δ¹⁸O_Sucrose), an eventuality that we minimized by extensive subsampling.

2.3 Gas exchange measurements

Leaf gas exchange measurements were performed during a 2-week interval between weeks 7 and 9 as described in Baca Cabrera et al. (2020). In brief, net assimilation (A, μmol CO₂ m⁻² s⁻¹), leaf transpiration (E_leaf, mmol H₂O m⁻² s⁻¹) and stomatal conductance to water vapour or CO₂ ($${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$, mol H₂O m⁻² s⁻¹ or $${g}_{s{{\rm{CO}}}_{2}}$$, mol CO₂ m⁻² s⁻¹) were measured on 6–12 plants per treatment with a LI-6400xt (Li-Cor) portable CO₂/H₂O gas exchange system with a clamp-on leaf cuvette (2 × 3 cm), installed in a separate plant growth chamber (E15, Conviron). For measurements, individual plants were removed from their growth chamber, and the midsection of the youngest fully developed leaf blades of four tillers was enclosed in the leaf cuvette. A, g_s and E_leaf were measured at a leaf temperature of 21°C (air temperature in the leaf cuvette ranged between c. 20.0–21.0°C) and a PPFD of 800 µmol m⁻² s⁻¹ and with CO₂ concentration and RH in the leaf cuvette (sample chamber) set equal to the conditions in the original growth environment.

Air for the leaf cuvette was supplied by mixing CO₂-free, dry air (with 21% O₂) and tank CO₂ (from Carbo Kohlensäurewerke or Linde AG, see Section 2.1) using mass flow controllers. Measurements were logged under steady-state conditions for stomatal conductance and water vapour concentration.

2.4 Mesophyll conductance to CO₂

2.5 Isotope analysis

Water samples were analyzed by cavity ring-down spectroscopy as described in Liu et al. (2016). A total of 1 µL of a water sample was injected into an A0211 high-precision vapourizer coupled to an L2110-i-CRDS (both Picarro Inc.). Each sample was measured 5–12 times depending on memory effects. After every 15–25 samples, heavy and light laboratory water standards, spanning the range of δ¹⁸O values in the data set and previously calibrated against V-SMOW, V-GISP and V-SLAP, were measured for SMOW-scaling and possible drift correction. Analytical uncertainty (standard deviation, SD) was <0.2‰ and <1.0‰ for δ¹⁸O and δ²H, respectively.

Sucrose samples were measured by isotope ratio mass spectrometry (IRMS) as in Baca Cabrera et al. (2021). Each sample was measured against a laboratory working standard carbon monoxide gas, previously calibrated against a secondary isotope standard (IAEA-601, accuracy of calibration ±0.25‰ SD). Solid internal laboratory standards (cotton powder) were run each time after the measurement of four samples for possible drift correction and for SMOW-scaling. The precision (SD) for the laboratory standard was <0.3‰.

We also checked for possible fractionation effects during HPLC sucrose separation, by comparing untreated fine ground sucrose directly weighed into silver cups with the same sucrose obtained after passage through the HPLC column. For the latter, we prepared a standard-mix (containing fructan, sucrose, glucose and fructose in typical tissue concentrations) and collected the sucrose fraction eluting from the HPLC column. No difference was detected for δ¹⁸O_Sucrose of the untreated sucrose (30.8 ± 0.7‰ SD) compared with the sucrose from the standard-mix passed through the HPLC (31.0 ± 0.3‰ SD).

Additionally, δ¹⁸O and δ²H of water vapour in the growth chambers (δ¹⁸O_Vapour/δ²H_Vapour) was measured by cavity ring-down spectroscopy as described in Liu et al. (2016). Here, we measured δ¹⁸O_Vapour/δ²H_Vapour continuously for 2 weeks when canopies were closed, both during the light and the dark periods. δ¹⁸O_Vapour was constant across experimental runs and treatments but was c. 1‰ more enriched during the dark period (−14.2‰ ± 0.5‰ SD) than during the light period (−15.2‰ ± 0.6‰ SD). A similar behaviour was observed for δ²H_Vapour (−104.6‰ ± 3.8‰ SD during the dark period and −109.6‰ ± 4.4‰ SD during the light period).

2.6 Evaporative site ¹⁸O enrichment

The ¹⁸O enrichment of evaporative sites water above source water (∆¹⁸O_e) during the light period was calculated using the Craig–Gordon model (Craig & Gordon, 1965; Dongmann et al., 1974) as presented in Cernusak et al. (2016) and given in Equation 1. w_a/w_i was calculated based on daytime air temperature (20°C) and RH (50% or 75%) inside the chambers, and canopy surface temperature, which was estimated using measurements of six type T thermocouples (in-house, custom-made) attached to mature leaf blades across each canopy. ε⁺ and ε_k were calculated as in Cernusak et al. (2016) (see Table 1), with stomatal conductance and boundary layer conductance (g_b) data obtained from leaf gas exchange measurements and rates of water loss from leaf replicas (Grace & Wilson, 1976) (g_b ≈ 0.2 mol m⁻² s⁻¹ for all treatments), respectively. Calculation of Δ¹⁸O_e—as given here—makes the (customary) simplifying assumptions that cuticular transpiration did not occur, stomatal conductance was nonpatchy and there were no gradients in leaf temperature, RH and stomatal conductance between the base and the top of the canopies.

2.7 The fractions of leaf water in nonphotosynthetic leaf tissue and its ¹⁸O enrichment

According to isotopic mass balance, Δ¹⁸O_LW is a mass-weighted average of the Δ¹⁸O of its photosynthetic (Δ¹⁸O_SSW) and nonphotosynthetic (Δ¹⁸O_non-SSW) components, with f_ssw and f_non-SSW the mass fractions of photosynthetic and nonphotosynthetic leaf water (f_non-SSW = 1 − f_SSW) (Equation 6). Δ¹⁸O_LW was obtained experimentally as explained above, and Δ¹⁸O_SSW was estimated as Δ¹⁸O_Sucrose − ε_bio, with ε_bio the ¹⁸O fractionation between carbonyl groups and water in which they are formed (Barbour, 2007). ε_bio (=26.7‰) was estimated from the temperature-dependence of ε_bio associated with cellulose synthesis in aquatic plants (Sternberg & Ellsworth, 2011; see also Hirl et al., 2021) and the nominal daytime air temperature in the growth chambers (including treatment-related offsets of canopy temperature in the calculation would not change the temperature-dependent ε_bio by more than ± 0.15‰). This estimate of ε_bio is also close to the (temperature-independent, constant) 27‰ which has been used most commonly in this type of analysis (Barbour, 2007; Lehmann et al., 2017).

To allow for and explore the uncertainty of this well-mixed mesophyll assumption, we also used higher values of f_non-SSW, which effectively treated part of the mesophyll (e.g., some or all the mesophyll vacuole volume) as a component of the nonmesophyll fraction, that is, reflecting a Δ¹⁸O equal to that of the nonmesophyll water (Δ¹⁸O_non-SSW). The most extreme scenario set f_non-SSW = 0.88, which would include the entire mesophyll vacuole volume in the nonmesophyll fraction. In this scenario, the photosynthetic leaf water fraction (f_SSW) equaled 0.12 and was represented solely by the combined water volume fraction of the mesophyll cytosol and chloroplasts. The latter was calculated from the volume fraction of cytosol plus chloroplasts (26%) in mesophyll cells of barley leaves (Winter et al., 1993) as 0.26 × (1 − 0.53). As an intermediate scenario we used f_non-SSW = 0.70. These higher estimates of f_non-SSW were then used to analyze the sensitivity of Δ¹⁸O_non-SSW (Equation 15), φ_non-SSW (=1  Δ¹⁸O_non-SSW/Δ¹⁸O_e) and effective path length (L; calculated by the Péclet effect model defined by Equation 7) to variation of f_non-SSW.

2.8 Data selection and statistics

In the first step, linear mixed models were fitted to test the effect of the diel period (day vs. night) on Δ¹⁸O_LW (n = 160) and Δ¹⁸O_SSW (n = 70). All available subsamples (pseudo-replicates) were included in the analysis, with the growth chamber and experimental run defined as the random factors. A small but significant diel trend between the end of the light period and the end of the dark period was detected for Δ¹⁸O_LW, but not for Δ¹⁸O_SSW (Supporting Information: Table S1). Therefore, we decided to: (1) pool day and night Δ¹⁸O_SSW data together; and (2) only use Δ¹⁸O_LW data obtained near the end of the light period for further calculations, under the assumptions that these values better reflect ¹⁸O enrichment processes during photosynthesis. Additionally, we calculated φ_LW and φ_non-SSW (which are derived from Δ¹⁸O_LW) using data obtained either near the end of the light or the dark period but observed only very small differences and no variation in the CO₂ and RH statistical effects on these variables (Supporting Information: Figure S1), indicating that our data selection was reasonable for further analyses.

In a second step, canopy scale replicates were obtained by pooling all individual subsamples from each canopy (n = 3–5) and two-way analysis of variance (ANOVA) tests used to assess the effects of CO₂, RH and their interaction on Δ¹⁸O_LW, Δ¹⁸O_SSW, Δ¹⁸O_e and all derived parameters. The same statistical analysis was used for the gas exchange parameters but based on individual leaf replicates (n = 6–12). Additionally, ordinary least-squares linear regressions were performed to test the relationship between the different parameters (using treatment averages). All statistical analyses were performed in R v.4.0.2 (R Core Team, 2022). The R packages nlme (Pinheiro et al., 2019) and ggplot2 (Wickham, 2016) were used for fitting linear mixed models and data plotting, respectively.

3.1 ¹⁸O enrichment of bulk leaf water (Δ¹⁸O_LW), photosynthetic tissue water (Δ¹⁸O_SSW), evaporative site enrichment (Δ¹⁸O_e) and relations among them

Δ¹⁸O_LW determined near the end of the 16 h-long periods of constant light ranged between 7.9‰ and 10.2‰ and responded significantly to [CO₂], with the increase of [CO₂] from 200 to 800 μmol mol⁻¹ causing decreases of Δ¹⁸O_LW of 1.3‰ at a daytime RH of 50% (designated ‘low RH’ in the following) and 2.3‰ at a daytime RH of 75% (‘high RH’) (Table 2, Supporting Information: Figure S2). RH alone and the interaction of RH and [CO₂] had no significant effects on Δ¹⁸O_LW. Similar treatment effects (or their absence) were detected for Δ¹⁸O_LW near the end of the dark period. However, Δ¹⁸O_LW changed slightly more between the end of the light and the end of the dark period in the low than the high RH treatment, likely owing to the fact that the low RH treatment involved a 25%-increase of RH between the light and dark periods, while RH was kept constant in the high RH treatment. Overall, however, the changes of Δ¹⁸O_LW between the end of the light and the end of the dark period were small (∼1.6‰ at low RH and ∼0.8‰ at high RH). These effects were consistent in replicated (n = 3–5) mesocosm-scale experiments. Although the absence of a significant daytime-RH effect on Δ¹⁸O_LW was unexpected, plots of δ¹⁸O versus δ²H—made using end-of-the-light-period leaf water and leaf growth zone samples—demonstrated a significantly steeper slope (Figure 1) and a lower deuterium deviation with reference to the global meteoric water line (Supporting Information: Table S2) for the high RH compared with the low RH treatment, for all [CO₂] levels, supporting a differential effect of daytime evaporative conditions on bulk leaf water isotope composition (Voelker et al., 2014) (see also discussion under Section 4.4).

The ¹⁸O enrichment of photosynthetic medium water (Δ¹⁸O_SSW) was calculated as Δ¹⁸O_Sucrose − ε_bio (with ε_bio constant at 26.7‰; see Section 2.7). Δ¹⁸O_Sucrose did not differ between samples collected between the end of the light and end of the dark period (Supporting Information: Table S1). Δ¹⁸O_SSW responded to both RH and [CO₂] treatments, with no significant interaction between RH and [CO₂] (Table 2, Supporting Information: Figure S2). That is, the RH response was the same for all CO₂ treatments. In all treatments, Δ¹⁸O_SSW was more enriched than Δ¹⁸O_LW, and this enrichment was much greater at low RH (+7.4 to +8.7‰) than at high RH (+2.2 to +2.4‰). At the same time, Δ¹⁸O_SSW decreased by ∼2.4‰ between 200 and 800 μmol mol⁻¹ CO₂. The average RH and [CO₂] sensitivities of Δ¹⁸O_SSW were –0.24‰/% and –0.004‰/μmol mol⁻¹.

Craig–Gordon modelled evaporative site enrichment (Δ¹⁸O_e) also responded significantly to RH and [CO₂] (Table 2). However, the [CO₂] effect on Δ¹⁸O_e was opposite to that for Δ¹⁸O_SSW, as Δ¹⁸O_e increased (by 2.1‰ at low RH and 2.8‰ at high RH) when [CO₂] during growth was increased from 200 to 800 μmol mol⁻¹. The average RH and CO₂ sensitivities of Δ¹⁸O_e were −0.28‰/% and +0.004‰/μmol mol⁻¹.

Δ¹⁸O_SSW was significantly related to Δ¹⁸O_e (R² = 0.69, p = 0.04), but not to Δ¹⁸O_LW (R² = 0.20, p = 0.38) if all treatments were pooled (Figure 2). Nevertheless, there were treatment-specific differences between Δ¹⁸O_SSW and Δ¹⁸O_e. While Δ¹⁸O_SSW was similar to Δ¹⁸O_e at ambient [CO₂], Δ¹⁸O_SSW was higher than Δ¹⁸O_e at low [CO₂] and lower at high [CO₂]. This [CO₂] effect on Δ¹⁸O_SSW−Δ¹⁸O_e was very similar in the two RH treatments (Figure 2b).

The proportional difference between Δ¹⁸O_LW and Δ¹⁸O_e (φ_LW = 1 − Δ¹⁸O_LW/Δ¹⁸O_e) varied between 0.03 and 0.57, and was influenced by RH and [CO₂], as well as their interaction (Table 2). φ_LW was higher at low RH than at high RH at every [CO₂] and increased with [CO₂], with this [CO₂]-driven effect being much higher at high RH. Conversely, the proportional difference between photosynthetic tissue water and evaporative site ¹⁸O enrichment (φ_SSW) was smaller (–0.17 to 0.23) and did not respond to RH, but increasing [CO₂] caused an increase of φ_SSW.

3.2 Gas exchange parameters, LWC and turnover

Leaf measurements of stomatal conductance for water vapour ($${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$) and transpiration (E_leaf) of paired plants were performed under conditions approximating those in the growth environment in each experimental run (Table 2) and have been reported by Baca Cabrera et al. (2020). Briefly, $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ responded significantly to [CO₂] and RH and their interaction. Specifically, $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ decreased exponentially with [CO₂] with an RH-sensitivity that decreased with [CO₂]. In the low [CO₂] treatment, $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ was 2.6-fold greater at high than at low RH, but only 1.5-fold greater at high [CO₂]). E_leaf also decreased exponentially (by ∼62%) with increasing [CO₂] and was higher at low than at high RH (+12%, +22% and +45% at low, ambient and high [CO₂]). Notably, the $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ to E_leaf ratio decreased with [CO₂]. This effect was (at least partially) related to systematic differences in air temperature among treatments (up to 1.3°C), caused by differences in E_leaf (leaf temperature was fixed at 21°C), associated offsets between leaf and air temperature (up to 0.9°C) (Figure 3) and offsets of actual relative to nominal %RH conditions (up to 3% RH) during gas exchange measurements in the leaf cuvette. We also observed systematic differences among treatments in the average canopy temperature (as measured with thermocouples; range 19.1–21.0°C), which were inversely related to canopy scale transpiration (Baca Cabrera et al., 2020). This indicates that the results observed at the leaf cuvette level were essentially also reflected at the entire sward scale.

Mesophyll conductance (g_m) decreased near-linearly by ~54% between low and high [CO₂] and was ~26% smaller at high than at low RH (Table 2). Also, g_m correlated linearly with E_leaf (R² = 0.89), but the correlation between g_m and g_s was not significant.

LWC was practically the same in the different treatments (12.2–12.9 mol m⁻²), except at high [CO₂]/low RH, where LWC was 11% higher than the average of the other treatments (Table 2). Accordingly, the turnover time of bulk leaf water (calculated as LWC/E_leaf) was primarily determined by the variation of E_leaf and varied between 62 min (low [CO₂] and low RH) and 212 min (high [CO₂] and high RH).

3.3 ¹⁸O enrichment of nonphotosynthetic tissue water

¹⁸O enrichment of the nonphotosynthetic leaf water fraction (Δ¹⁸O_non-SSW) was first estimated by Equation 15 using a f_non-SSW = 0.53, the anatomically based estimate of the nonmesophyll water fraction in the bulk leaf (Figure 4a, left panel). The proportional offset of Δ¹⁸O_non-SSW from Δ¹⁸O_e (φ_non-SSW = 1 − Δ¹⁸O_non-SSW/Δ¹⁸O_e) is given in Figure 4b (left panel). The estimates of Δ¹⁸O_non-SSW thus obtained relied on the assumption that the mesophyll was well-mixed (as explained in Section 2.7, and consistent with Equation 7), which implied that Δ¹⁸O_SSW was representative for the entire mesophyll, including the mesophyll vacuoles. Under this assumption, Δ¹⁸O_non-SSW was 2.0‰ on average for the low RH treatments, implying that the δ¹⁸O of water in the non-SSW fraction was relatively close to source water (specifically, decreasing f_non-SSW below 0.48 predicted negative Δ¹⁸O_non-SSW for these treatments). Conversely, at high RH, isotopic mass balance (Equation 15) indicated that the nonphotosynthetic leaf water fraction was distinctly ¹⁸O enriched, and particularly so at low [CO₂] (+8.3‰).

Increasing f_non-SSW from 0.53 to 0.88—which considered nonperfect mixing of the mesophyll, that is, an increasing proportion of the mesophyll having the same Δ¹⁸O as the nonmesophyll tissue—predicted increases of Δ¹⁸O_non-SSW (Figure 4a) and corresponding decreases of φ_non-SSW in all treatments (Figure 4b). This effect was most apparent at low RH. That is, with the type of sensitivity analysis used here, consideration of nonperfect mixing of the mesophyll was associated with a much smaller effect on estimates of Δ¹⁸O_non-SSW at low RH than at high RH.

3.4 Relationships between gas exchange parameters and ¹⁸O enrichment of whole-leaf water, and of the photosynthetic and nonphotosynthetic tissue water fractions

Overall, the relationship between E_leaf and φ_LW was nonsignificant (R² = 0.14; p > 0.05) and exhibited great scatter (Figure 5a). However, this scatter hid systematic, but opposing RH and [CO₂] effects on the relationship between E_leaf and φ_LW: an RH-driven positive relationship evident at each individual [CO₂] level, and a counteracting negative relationship driven by variation of [CO₂], which was present at both RH levels. Meanwhile, E_leaf and φ_SSW (calculated as φ_SSW = 1 − Δ¹⁸O_SSW/Δ¹⁸O_e) correlated negatively across the treatments (R² = 0.81; p < 0.05) (Figure 5b). Conversely, the relationship between E_leaf and φ_non-SSW (calculated as φ_SSW = 1 − Δ¹⁸O_non-SSW/Δ¹⁸O_e) was nonsignificant irrespective of the value at which f_non-SSW was fixed (Figure 5c for f_non-SSW = 0.53; data for other values of f_non-SSW not shown). Also, φ_non-SSW displayed qualitatively similar treatment-dependent relationships with E_leaf as φ_LW (compare Figure 5a,c).

φ_SSW was negatively correlated with $${g}_{s{{\rm{CO}}}_{2}}$$, g_m and total conductance (g_total, calculated as 1/(1/$${g}_{s{{\rm{CO}}}_{2}}$$ + 1/g_m) (Figure 6). The correlation between φ_SSW and g_total was particularly strong (R² = 0.93; p < 0.01), while the one with $${g}_{s{{\rm{CO}}}_{2}}$$ alone (R² = 0.79; p = 0.02) appeared to vary between RH levels. The correlation of φ_SSW with g_m was only marginally significant (R² = 0.62; p = 0.06). In contrast, φ_LW and φ_non-SSW did not correlate with g_m (not shown), similar to the relationships with E_leaf, which were also nonsignificant (compare with Figure 5a,c).

φ_non-SSW demonstrated a very close, saturating relationship with the ratio of E_leaf to water vapour influx, estimated as $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ × w_a (Pseudo-R² = 0.99, p < 0.05 for fixed f_non-SSW of 0.53) (Figure 7). Increasing f_non-SSW from 0.53 to 0.70 or 0.88 did not alter this relationship in a qualitative sense (Pseudo-R² = 0.98 and 0.96, respectively), although it did change the value of the asymptote. Similar, although more blurred relationships existed between φ_LW and the ratio of E_leaf to water vapour influx (not shown). Conversely, there was no significant relationship (p > 0.05) between φ_SSW and the ratio of E_leaf to water vapour influx.

Using the data presented in Table 2 to fit the Péclet model as presented in Equation 7, we obtained estimates of the effective path length (termed L_CG) for the different treatments (Figure 4c). These estimates of L were sensitive to (i) the assumptions of f_non-SSW used in the estimation of Δ¹⁸O_non-SSW, and growth conditions of (ii) [CO₂] and (iii) daytime RH (Figure 4c). In general, however, L increased with increasing [CO₂] and decreased with daytime RH. Specifically, at f_non-SSW = 0.53, L increased from 63 to 132 mm between low and ambient [CO₂] (no value could be fitted for the high [CO₂] level) at 50% RH and from 4 to 182 mm between high and low [CO₂] at 75% RH. Alternatively, we also estimated L (termed L_SSW, Figure 4d) using the same data, except for replacing Δ¹⁸O_e by Δ¹⁸O_SSW, the Δ¹⁸O of the photosynthetic water pool, which is represented by the first term on the right-hand side of Equation 7. This yielded qualitatively similar estimates of L, except that the effect of [CO₂] on L was attenuated. Thus, estimates of L_SSW were higher than L_CG at low [CO₂], but lower at high [CO₂].

4.1 On the approach—Advances and caution

Here, we present a new type of compartmental analysis of bulk leaf water ¹⁸O enrichment (Δ¹⁸O_LW), based on the propositions of Holloway-Phillips et al. (2016) (see their eqns. 9–11). This analysis distinguished Δ¹⁸O in the photosynthetic tissue (Δ¹⁸O_SSW) and nonphotosynthetic tissue (Δ¹⁸O_non-SSW) by using an isotopic mass balance (Equation 6) and estimates of the bulk leaf water fractions in photosynthetic and nonphotosynthetic leaf tissues, where the latter includes the veins and epidermis. This analysis rested partly, but significantly, on published comprehensive leaf anatomical analyses of C₃ grasses (Dengler et al., 1994), including L. perenne (Charles-Edwards et al., 1974), the object of this work. This approach is functionally distinct and alternative to previous works which divided bulk leaf water into leaf lamina and primary vein water (see Section 1, Gan et al., 2002, 2003; Yakir et al., 1994) particularly for how water in the epidermis (a co-dominant component of bulk leaf water) is attributed to the functional parts of the leaf: exclusively associated with the nonphotosynthetic leaf tissue (which also includes the veins) in the present approach; and mostly with the lamina fraction (which includes the photosynthetic tissue) when the bulk leaf is separated in its lamina and (main) vein parts, in the earlier approach. Likewise, consistent with the proposition of Holloway-Phillips et al. (2016), we argue that the photosynthetic tissue was in (or near) equilibrium with ¹⁸O-enrichment at the evaporative sites as predicted by the Craig–Gordon model (although small offsets were systematic; see discussion under Section 4.2). Conversely, the nonphotosynthetic tissue, including vascular tissue and epidermis, appeared to display a treatment-dependent Péclet effect (Section 4.2).

However, the present analysis relied on several assumptions, including (1) that ε_bio for sucrose was the same (26.7‰) in all treatments, meaning that there was a constant offset between Δ¹⁸O_Sucrose and Δ¹⁸O_SSW across treatments. Assumptions specifically important for the analysis of nonphotosynthetic tissue water fraction were more numerous and required, for all treatments, that (2) the mesophyll was well-mixed, (3) f_non-SSW was 0.53, (4) the ratio of assimilation to transpiration (A/E) was constant over the leaf surface and (5) that Equation 7, which was used to calculate L in the nonphotosynthetic tissue water fraction, provides a reasonable description of Δ¹⁸O_LW. Considering uncertainty, however, we did relax specific assumptions of the analysis, as we discuss below.

Concerning (1): the temperature-dependent (Hirl et al., 2021; Sternberg & Ellsworth, 2011) ε_bio chosen here was very close to the ‘constant 27‰’ assumption which has been used most commonly (Barbour, 2007; Cernusak et al., 2005; Lehmann et al., 2017) as we used a thermal environment for which the ε_bio estimates of the two approaches converged closely (air temperature in the light period = 20°C). This estimate of ε_bio is taken from that for cellulose formation (Barbour, 2007; Cernusak et al., 2003); we are not aware of independent confirmations of ε_bio for sucrose in photosynthesizing leaves. Importantly, however, we found no evidence for incomplete equilibration of carbonyl groups with sucrose synthesis water, as the δ¹⁸O of sucrose did not differ significantly between plants grown in the presence of ¹⁸O-enriched or ¹⁸O-depleted CO₂ (Supporting Information: Figure S3), similarly to cellulose (Liu et al., 2016). Also, the δ¹⁸O of bulk leaf water was not affected by the ¹⁸O of CO₂ (p = 0.7).

Further, (2): Δ¹⁸O_Sucrose did not vary between the end of the light period and the end of the dark period, indicating that diurnal variation in mobilization of sucrose from the vacuole or recycling of sucrose via hydrolysis or fructan metabolism (Lattanzi et al., 2012; Pollock & Cairns, 1991) did not alter the ¹⁸O composition of sucrose. The apparent constancy of Δ¹⁸O_Sucrose could be related to a relative constancy of Δ¹⁸O of mesophyll water, as we observed only very small (if any) variation of Δ¹⁸O_LW between the end of the day and end of the night. Conversely, however, Lehmann et al. (2017) found that hexoses were depleted by ∼2‰ relative to sucrose, equivalent to a ∼2‰ depletion of the synthesis water of hexoses relative to sucrose, if the ε_bio was the same for both sugars. This could indicate that metabolism of hexoses derived from sucrose was associated with significant oxygen exchange in a less ¹⁸O-enriched environment locally, or that the hexose-to-sucrose ratio varied along the leaf in parallel with ¹⁸O enrichment of medium water (Lehmann et al., 2017). Evidence for a different subcellular localization was presented by Wagner et al. (1983) who found hexoses to be virtually exclusively localized in the vacuole and a significant fraction of the sucrose in the cytosol of mesophyll cells isolated from barley primary leaves. Most interestingly, Koroleva et al. (1998) found a much higher hexose-to-sucrose ratio in epidermal cells of barley leaves than in the mesophyll (or bundle sheath parenchyma). If epidermis water is less ¹⁸O enriched than mesophyll water (as we suggest), this could also explain a lesser ¹⁸O enrichment of hexoses in whole-leaf extracts.

Nevertheless, we realize that direct empirical proof for the (eventual) constancy (or absence thereof) of ε_bio for leaf sucrose (ε_{bio Sucrose}) is presently missing. Certainly, there is a critical need for experimental determinations across a range of plant functional groups, environmental conditions and diurnal cycles (Holloway–Phillips et al. 2022). This includes possible metabolically based variation of ε_{bio Sucrose} (a) at the site of primary synthesis in the mesophyll, (b) during transport to and into sieve elements, and (c) in association with sucrose-consumption and resynthesis during eventual storage along the path in nonphotosynthetic leaf tissue.

Although we did not perform anatomical investigations ourselves, we believe that assumption (3) was well-founded for a f_non-SSW = 0.53, if the mesophyll was well-mixed. Estimates of mesophyll and nonmesophyll proportions of C₃ grass leaves are known to vary comparatively little among and within studies (e.g., Charles-Edwards et al., 1974; Dengler et al., 1994; Garnier & Laurent, 1994; Winter et al., 1993). In particular, fractions of mesophyll and nonmesophyll tissue observed by Charles-Edwards et al. (1974) were remarkably constant among genotypes of L. perenne across contrasting growth conditions. Although leaf thickness correlated with the proportion of mesophyll, this parameter displayed little variation across diverse environments (Charles-Edwards et al., 1974). Also, we found very little variation of leaf thickness, based on LWC (Table 2). Moreover, leaf length and width, and epidermal cell number and length did not differ between treatments (Baca Cabrera et al., 2020). Finally, a minimum f_non-SSW ~ 0.5 was also supported by isotopic mass balance (Equation 15). Thus, we propose as the most parsimonious hypothesis that the proportions of mesophyll (and nonmesophyll) tissue, and hence f_SSW (and f_non-SSW) were very closely similar between the different growth environments.

Assumption (4) is predicted by optimal stomatal control theory (Cowan & Farquhar, 1977) and supported by the close agreement between Δ¹⁸O_e and Δ¹⁸O_SSW for the ambient [CO₂] treatments at both high and low RH. Yet, we did observe small but systematic [CO₂]-dependent variation of φ_SSW (range –0.17 to +0.23), as we discuss below.

4.2 ¹⁸O enrichment of sucrose more closely related to evaporative site enrichment than to whole-leaf water

Our work confirms the observation of Lehmann et al. (2017) of a significant and RH-dependent underestimation of Δ¹⁸O_Sucrose by Δ¹⁸O_LW + ε_bio (26.7‰) in L. perenne. In fact, the underestimation observed at low daytime RH was higher than that observed by Lehmann et al. (2017). Thus, the RH sensitivity of Δ¹⁸O_Sucrose (−0.24‰/%) was markedly greater at every [CO₂] level than the RH sensitivity estimated for their data (ca. −0.16‰/% on average of the two grasses).

Additionally, our data reveal an overall closer agreement of Δ¹⁸O_SSW (i.e., Δ¹⁸O_Sucrose − ε_bio) with Δ¹⁸O_e than with Δ¹⁸O_LW. This result differs from the conclusion of Cernusak et al. (2003) which was based on the comparison of the ¹⁸O enrichment of phloem dry matter (mostly sucrose; e.g., Smith & Milburn, 1980) with lamina water (which excluded the main veins) or Δ¹⁸O_e in a dicot (R. communis). However, as our analysis was based on bulk leaf water, it is not strictly comparable with that of Cernusak et al. (2003). As shown by Gan et al. (2003) and Barbour et al. (2021), accounting for the vein fraction—which represents by itself only a fraction of the total nonphotosynthetic tissue water (see above)—already reduces substantially the discrepancy between ¹⁸O enrichment of lamina water and Δ¹⁸O_e. Unfortunately, sufficiently-rapid, artifact-free physical separation of the nonphotosynthetic and photosynthetic leaf tissue of L. perenne (or other C₃ grasses) appears to be practically impossible at present (see also discussions in Cernusak et al., 2003).

The offset between Δ¹⁸O_SSW and Δ¹⁸O_e was influenced strongly by [CO₂], but not by RH or the interaction of [CO₂] and RH. This effect correlated closely with the effect of [CO₂] on E_leaf, g_s, g_m and g_total. To our best knowledge, the relationship between φ_SSW and gas exchange parameters has not been investigated previously, reducing opportunities for discussion. However, Ferrio et al. (2012) observed an apparent coordination between mesophyll (and hydraulic) conductance and effective path length (L) in vein-severing experiments with droughted and control plants of Vitis vinifera. Our observation of a negative relationship between E_leaf and φ_SSW (Figure 5b) also implies a (strong) decrease of L with increasing E_leaf, an effect previously observed in tree species (Loucos et al., 2015; Song et al., 2013); and—since E_leaf and g_m (as well as g_total) were closely correlated—there may have also been a negative association between g_m and L, consistent with observations of Ferrio et al. (2012), when applied only to photosynthetic medium water in the present case.

An increase in L under increasing [CO₂] for both RH levels could also be related to changes in anatomical features of stomata. Studies in Arabidopsis (Larcher et al., 2015) and mangrove plants (Sternberg & Manganiello, 2014) indicated a decrease in L with increasing stomatal density, associated with a decrease in the distance between the stomatal pores and the veins and an increase in the total cross-sectional area through which mesophyll water flows. As [CO₂] has risen, stomatal density has decreased (and pore size increased) over geological time (Franks & Beerling, 2009). A similar behaviour has been observed in controlled experiments under elevated [CO₂] for a wide variety of species and accessions of Arabidopsis thaliana (Hetherington & Woodward, 2003). However, metaanalyses of FACE experiments indicate a weaker relationship between [CO₂] and stomatal density, with low consistency (Ainsworth & Rogers, 2007; Poorter et al., 2022). We did not measure stomatal density or pore size in our experiment, but the observed positive relationship between [CO₂] and L could be indicating a stomatal density-driven effect (i.e., increase in L with increasing CO₂, due to lower stomatal density). Certainly, disentangling the effect of stomatal density/pore size and stomatal aperture (gas exchange) on L under rising [CO₂] should be a target for future research.

Notably, based on our analysis, Δ¹⁸O_SSW was slightly greater than Δ¹⁸O_e (implying that φ_SSW < 0) at low [CO₂] in both RH treatments. As Δ¹⁸O_SSW has not been analyzed previously, there is no other work with which this observation can be compared. However, negative values of φ_LW have been observed quite frequently in different plant functional groups (Cernusak et al., 2016, 2022), including in grasses (Helliker & Ehleringer, 2000), although such relationships are theoretically unexpected in steady-state conditions (Farquhar & Gan, 2003; Gan et al., 2003; Ogée et al., 2007). Because of the expectation that Δ¹⁸O_SSW should always be greater than Δ¹⁸O_LW, we would assume that such reported negative φ_LW values were also associated with (even more) negative φ_SSW values.

Errors in the simplifying assumptions for the calculation of Δ¹⁸O_e (i.e., absence of cuticular transpiration, stomatal patchiness and RH and temperature gradients between base and tip of the leaf and the canopy) could perhaps also contribute to a mismatch between Δ¹⁸O_e and Δ¹⁸O_SSW. In particular, the increase in Δ¹⁸O_e from the leaf base toward the tip, which has been observed in grasses (Helliker & Ehleringer, 2000; Ogée et al., 2007) was not included in our model. A similar increase is expected for Δ¹⁸O_SSW but has not been studied so far. At the same time, transpiration (E) (Helliker & Ehleringer, 2000; Ogée et al., 2007) and photosynthesis (A) (Meinzer & Saliendra 1997; Xiong et al., 2015) are known to increase from the leaf base to the leaf tip in grasses, and relative changes of the A/E ratio may vary along the leaf (Ocheltree et al., 2012), deviating from the constant A/E assumption. To which degree this relationship is affected by [CO₂] has not been explored but could perhaps play a role in the spatial mismatch. Any spatial mismatch between (transpiration-weighted) Δ¹⁸O_e and (assimilation-weighted) Δ¹⁸O_SSW, with weighted maximum assimilation occurring closer to the leaf tip than transpiration (Lehmann et al., 2017), could perhaps explain why Δ¹⁸O_SSW > Δ¹⁸O_e or φ_SSW < 0 at low [CO₂] (see also Ogée et al., 2007). In the same sense, the opposite spatial pattern could explain why Δ¹⁸O_SSW < Δ¹⁸O_e or φ_SSW > 0 at high [CO₂], that is, the weighted maximum assimilation occurring further away from the leaf tip than transpiration. Also, the [CO₂]-dependent shifts could be connected to the fact that low [CO₂] caused high g_s and g_m, while the opposite may have been the case at high [CO₂]. Additionally, one may wonder, if a spatial mismatch between transpiration-weighted Δ¹⁸O_e and assimilation-weighted Δ¹⁸O_SSW could also be associated with variation of transpiration and photosynthetic activity with depth inside the leaf. Finally, the assumption of a constant ε_bio discussed in Section 4.1 could also play a role in the discrepancy between Δ¹⁸O_e and Δ¹⁸O_SSW at low and high [CO₂]. Clearly, these are important questions for future investigations.

4.3 ¹⁸O enrichment of nonphotosynthetic tissue water was sensitive to water vapour influx

Probably the greatest uncertainty in our analysis is connected with the estimation of Δ¹⁸O_non-SSW (and φ_non-SSW) as this was determined from the residual of Equation 6 (hence integrated virtually all uncertainties discussed above, see Equations 15) and Equation 5 depended on the adequacy of Equation 7 for estimation of L. For the latter, we believe that the (variable) discrepancy between Δ¹⁸O_SSW and Δ¹⁸O_LW, the comparatively close relationship between Δ¹⁸O_SSW and Δ¹⁸O_e, and the high φ_non-SSW at low RH support well the model represented by Equation 7. Also, we did account for uncertainty in the assumption that the mesophyll was well mixed, by increasing the proportion f_non-SSW in a sensitivity analysis. All results demonstrated that estimates of L, and relationships of Δ¹⁸O_non-SSW and φ_non-SSW with treatment effects and physiological parameters, remained the same in a qualitative sense.

In their analysis of the relative contributions of vapour and liquid water transport from the veins toward the stomata, Rockwell et al. (2014) predicted that an increase in E_leaf due to a decrease in ambient vapour mole fraction (w_a) would pull the distribution of leaf internal evaporation away from the perivascular space toward the stomata. Such an effect should drive an increase of L (Barbour et al., 2017). This expectation is consistent with the RH effect on L of the non-SSW leaf water fraction as estimated here, either by Equation 7 (Figure 4c) or by using a modified Equation 7, in which Δ¹⁸O_e was replaced by Δ¹⁸O_SSW in the first term on the right-hand side of the equation (Figure 4d). In both cases, the increase of L resulting from the decrease of RH in the growth environment was strong at every [CO₂] level and for every f_non-SSW used in the estimation of Δ¹⁸O_non-SSW.

At the same time, L increased pronouncedly with increasing [CO₂] in the growth environment. This effect was associated with a decrease in E_leaf, at both RH levels, independently of assumptions of f_non-SSW. Although Rockwell et al. (2014) did not explore effects of different [CO₂] levels, this result is consistent with pulling the (evaporation-weighted) site of transpiration toward the periveinal/perivascular space when [CO₂] is decreased. The leaf gas exchange measurements—performed under environmental conditions approximating those of the growth environment—indicated a very strong [CO₂] effect on $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ which entrained a parallel effect on E_leaf, the temperature gradient between leaf (T_L) and air (T_a), T_L − T_a, and hence w_i − w_a inside the leaf cuvette at both RH levels (Figure 3). Clearly, [CO₂] must have had very strong effects on the energy balance of leaves, increasing the ratio of sensible to latent heat transfer, a situation that would cause an increase in the fraction of peristomatal evaporation (Rockwell et al., 2014). Again, this interpretation is consistent with our estimates of L increasing with [CO₂]. The values of L (termed L_CG) based on Equation 7 cover a wide range (2–182 mm for a f_non-SSW 0.53). The range of L estimated with Δ¹⁸O_e replaced by Δ¹⁸O_SSW (termed L_SSW) was distinctly smaller (by about half) but displayed qualitatively the same treatment response. Interestingly, the ranges of L estimated here are similar in magnitude to the ones obtained for whole leaves of another C₃ grass (wheat) based on anatomical investigations and a range of scenarios concerning the site of evaporation within leaves (Barbour & Farquhar, 2004; Barbour et al., 2017).

An intriguing observation made here concerns the (saturating) relationship between φ_non-SSW and the ratio of E_leaf to water vapour influx (Figure 7). An analogous asymptotic relationship was not observed for φ_SSW but existed also for φ_LW (Pseudo-R² = 0.94), although the latter comprised more scatter due to the inclusion of the photosynthetic medium fraction in whole-leaf water. E_leaf equals source water influx into the leaf in the steady state, which very likely existed at the time of sampling (~14 h into the light period under constant environmental conditions, including irradiance). The ratio of E_leaf to water vapour influx abbreviates to (w_i – w_a)/w_a when $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$ is eliminated from both the numerator and denominator of the ratio. This shows that at a given RH, variation of w_i resulting from the variation of leaf temperature (controlled by $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$) was the main determinant of the variation of (w_i – w_a), as w_a was kept near constant in the growth chambers. At high RH, the variation of w_i drove a strong variation of φ_non-SSW, with (w_i – w_a)/w_a increasing dramatically in response to increasing [CO₂] due to the increase of T_L, which was a consequence of the g_s-driven decrease in E_leaf. Thus, at high RH, variation of φ_non-SSW appeared to be critically dependent on the energy balance of the leaf as controlled by $${g}_{s{{\rm{H}}}_{2}{\rm{O}}}$$. In that, it seems possible that the CO₂-dependent relationship between (w_i – w_a)/w_a and φ_non-SSW is somehow connected with the position (depth inside the leaf) of evaporative sites, and related effects on L, particularly at high humidity.

4.4 Absence of a daytime RH effect on ¹⁸O enrichment of leaf water—Unreal, uncommon or just unnoted and underexplored?

Commonly, Δ¹⁸O_LW exhibits a negative trend with RH (e.g., Cernusak et al., 2016, 2022; Helliker & Ehleringer, 2002; Hirl et al., 2019; Lehmann et al., 2017; Liu et al., 2017), a relation that we did not find here. But, we did find evidence of some effect of daytime evaporative conditions in bulk leaf water composition, based on the observations that the deuterium deviation with reference to the global meteoric water line (Voelker et al., 2014; Wei & Lee, 2019) was higher at low RH than at high RH, for all [CO₂] levels (Figure 1, Supporting Information: Table S2).

The commonly expected relation between RH and Δ¹⁸O_LW is generally interpreted in terms of predictions from the Craig–Gordon model (e.g., Roden & Ehleringer, 1999), which dictates a strong negative effect of increasing RH on Δ¹⁸O_e (Equation 1) and propagation of this enrichment in leaf water (Barbour, 2007; Cernusak et al., 2016, 2022; Craig & Gordon, 1965; Dongmann et al., 1974; Farquhar et al., 1989, 2007; Flanagan et al., 1991). However, a recent compilation of published and previously unpublished leaf water isotope data (data set S1 in Cernusak et al., 2022) demonstrates an (RH-insensitive) global average φ_LW of 0.16, which indicates that the RH sensitivity of Δ¹⁸O_LW should be about 16% smaller than that of Δ¹⁸O_e. However, φ_LW can be much greater (Cernusak et al., 2016), which would further reduce the RH sensitivity of Δ¹⁸O_LW in these cases. Specifically, the grassland (Hirl et al., 2019) and grass species (Stipa capillata, Zhao et al., 2014) included in the Cernusak et al. (2022) data set had an average φ_LW of 0.36 and 0.45, respectively, suggesting per se a 36% and 45% reduced RH sensitivity of Δ¹⁸O_LW relative to Δ¹⁸O_e. Moreover, φ_LW exhibited a common negative RH response of −0.005%⁻¹ in the studies of Hirl et al. (2019) and Zhao et al. (2014) (data collected between 10 AM and 4 PM), respectively (Supporting Information: Figure S4), a fact not highlighted or discussed before. This negative RH response of φ_LW implied an additional 13% reduction of the RH effect on Δ¹⁸O_LW in the range from 50% to 75% RH. In the present work, we observed a mean φ_LW of 0.37 and an RH-sensitivity of φ_LW (−0.01%⁻¹) which was twice that of Hirl et al. (2019) and Zhao et al. (2014). This greater negative RH response of φ_LW effectively cancelled the RH effect on Δ¹⁸O_LW in our investigations and was mainly related to the opposite effects of RH on Δ¹⁸O_non-SSW relative to Δ¹⁸O_SSW. Notably, other datasets with a nonsignificant RH response of Δ¹⁸O_LW are also included in the data set of Cernusak et al. (2022) and comprise the S. capillata data of Zhao et al. (2014; samples collected between 10 AM and 4 PM and reflecting an RH range from 26% to 56%). Notably, the RH sensitivity of ∆¹⁸O_LW which we calculate from the data of Lehmann et al. (2017) is also quite low, when compared to the average response of the global data set of Cernusak et al. (2022): −0.09‰/% versus −0.30‰/%. These findings and considerations all support the view that the magnitude and range of the φ_LW and ∆¹⁸O_LW data and their response to RH, as observed in the present work, fall inside the range of observations made previously by others in diverse environments (Cernusak et al., 2022).