Plant response to decreasing soil moisture under rising atmospheric CO₂ levels

Abstract

Widespread drought driven by global warming is predicted across the 21^st century, just as CO₂ level is projected to as much as double over the same time period. However, the potential interplay between increasing CO₂ and decreasing soil moisture on plant function is uncertain, as previous works have not successfully separated and quantified these two competing effects. Here we evaluated the interaction between soil moisture and CO₂ using stable carbon isotope measurements of Arabidopsis grown under the simultaneous modulation of an exhaustive range of both variables. Results showed that both increasing soil water content and increasing CO₂ increased isotopic discrimination, and that the soil moisture effect was not influenced by the CO₂ level and vice versa. Our data confirm that changes in the carbon isotope record of terrestrial organic matter preserved within the geologic record are best interpreted as deriving from changes in atmospheric CO₂.

Introduction

Because stomatal conductance at the leaf surface is responsive to soil moisture^1,2, workers have long used measurements of plant carbon isotope value to interpret plant performance (i.e., carbon assimilation) under drought in agricultural^3,4, ecological^5,6, and climate change studies^7,8,9. More recently, rapid rates of CO₂ rise due to the burning of fossils fuels have led researchers to examine the effect of increased CO₂ input on the net carbon isotope discrimination (Δ¹³C) imparted during photosynthesis, and have identified a significant change in fractionation across the levels of CO₂ rise expected during the next century (e.g. refs. ^10,11,12,13).

While multiple studies have evaluated the effect of soil moisture on Δ¹³C value under constant CO₂ (e.g. refs. ^14,15), and other studies have measured the effect of CO₂ on Δ¹³C value under constant soil moisture^10,12,16,17, experiments able to quantify the relative influence of both variables have remained elusive. Such experiments are critical for accurate interpretation of plant carbon assimilation from Δ¹³C value, particularly across changing time periods of rapidly changing CO₂. Here we report results from Arabidopsis plants grown in controlled experiments that varied both soil moisture and CO₂ systematically across a wide range, in order to quantify the influence of both variables on Δ¹³C value, and test for any interaction between the two.

Modeling approaches to evaluating the effect of water availability on plant response have focused on the examination of vapor pressure deficit (VPD) as the primary environmental driver (e.g. refs. ^18,19). For our experiments, we chose to hold relative humidity constant throughout our experiments, in order to isolate the effects of soil moisture, rather than VPD. We chose this focus because although VPD and soil moisture do interact and are coupled²⁰, soil water content has been shown to be the dominant driver of moisture stress within most ecosystems^21,22,23. In addition, numerical approaches have confirmed the importance of soil moisture on stomatal function and called for models to include better characterizations of soil hydrology²⁴.

We have chosen to analyze the response of biomass and Δ¹³C to CO₂ in terms of a non-linear function as the hyperbolic pattern is a fundamental feature of biological systems²⁵. Hyperbolic analysis is an established method for the analysis of CO₂ fertilization of plant biomass (e.g. refs. ^26,27). For the same reason, we also modeled the Δ¹³C response using a two-parameter rectangular hyperbola as within our previous work¹⁰.

$${\Delta }^{13}{{{\rm{C}}}}=\left[{AB}\left({{{{\rm{CO}}}}}_{2}+C\right)\right]/\left[A+B\left({{{{\rm{CO}}}}}_{2}+C\right)\right]$$

(1)

where A is the asymptote, or maximum fractionation for a given water treatment at infinite CO₂ (similar to 10⁶ ppm); B is a measure of responsiveness, where higher values for B indicate the effect of CO₂ on Δ¹³C saturates at lower CO₂ (i.e., the function reaches the asymptote quickly); and C is an offset such that Δ¹³C = 4.4‰ at CO₂ = 0 ppm (after²⁸). Fitted estimates of A and B were obtained by iterative optimization to minimize the sum of the residuals squared.

Results

Biomass

Figure 1 presents the CO₂ level provided for growth versus the above-ground biomass of harvested plants, separated by degree of water-stress as reflected by the gravimetric water content of the soil (i.e., θ_m = 27, 30, 45, 60%). Because biomass is not normally distributed across individual plants, we reported the data in terms of ln-transformed milligrams according to convention (after²⁹). Our measurements showed that biomass increased with increasing CO₂ at each level of θ_m (Fig. 1). Responsiveness to CO₂ level, as quantified by the slope parameter (B*) was remarkably similar across all levels of θ_m (i.e., within 0.04 units) (Table 1). Biomass increased with increasing θ_m value in a strongly non-linear fashion: between the θ_m values of 30% and 45%, biomass (mg) increased ~160% on average; between the θ_m values of 45% and 60%, biomass (mg) increased by ~40% (Fig. 2, Table 2).

Fig. 1: The effect of CO₂ and soil water content (θ_m) on above-ground biomass.

Data are plotted as the geometric mean value (±1σ) (after ref. ²⁹), and modeled using a hyperbolic function (after refs. ^26,27): Biomass_60% = [(4.9)(0.05)(CO₂ + 100)]/[4.9 + (0.05)(pCO₂ + 100)]; Biomass_45% = [(4.6)(0.04)(CO₂ + 100)]/[4.6 + (0.04)(pCO₂ + 100)]; Biomass_30% = [(3.7)(0.02)(CO₂ + 100)]/[3.7 + (0.02)(pCO₂ + 100)]; Biomass_27% = [(3.1)(0.06)(CO₂ + 100)]/[3.1 + (0.06)(pCO₂ + 100)].

Table 1 Biomass response curve fitting parameters A* and B* in response to increasing CO₂ level (data plotted in Fig. 1), where A* is the asymptote and B* is the slope at y = x = 100 ppm (after refs. ^26,27)

Fig. 2: The effect of soil water content (θ_m) on above-ground biomass at each level of CO₂.

Data are plotted as the mean value ± 1σ.

Table 2 Average biomass and stable carbon isotope (δ¹³C_p, δ¹³C_a, and Δ¹³C) data for each of the fifteen experiments performed

Some variation in biomass between individual plants grown under identical conditions is to be expected, as quantified by ref. ²⁹. who reported biomass variability from a wide range of growth experiments (n = 569). They found a median σ_lnw (standard deviation of ln-transformed biomass) value of 0.3; the average σ_lnw in our experiments ≤0.3 for all soil moisture treatments, indicating a typical level of variation between the individuals of our study.

Carbon isotopes

Stable carbon isotope measurements of plant tissues showed an increase in Δ¹³C with increasing CO₂ at each level of θ_m (Fig. 3). While the hyperbolic pattern of the Δ¹³C response to CO₂ fit very well (mean absolute error ≤0.4‰) for each value of θ_m (Fig. 3), we observed a marked difference in the absolute Δ¹³C value across θ_m treatments less than 45%. Specifically, we observed a consistent offset in Δ¹³C of 2.7 ± 0.3‰ between θ_m treatments of 30% and 45% and an additional 3.1 ± 0.3‰ offset between θ_m treatments of 27% and 30%, reflecting the known effect of water-stress on δ¹³C_p value (e.g. ref. ³⁰). However, when comparing the θ_m treatments ≥45%, the measured Δ¹³C value was nearly the same for both θ_m = 45% and 60% at all levels of CO₂ tested (i.e., the average difference was only 0.2‰).

Fig. 3: The combined effect of CO₂ concentration and soil water content (θ_m) on Δ¹³C value.

For each water treatment experiment (θ_m = 27, 30, 45, and 60%), the Δ¹³C value (data reported as mean ± 1σ, n = 10 to 12 plants) increased with increasing CO₂ following a general hyperbolic relationship (after Eq. (1)): Δ¹³C_60% = [(28.6)(0.32)(CO₂ + 16)]/[28.6 + (0.32)(pCO₂ + 16)]; Δ¹³C_45% = [(28.3)(0.33)(CO₂ + 16)]/[28.3 + (0.33)(pCO₂ + 16)]; Δ¹³C_30% = [(27.3)(0.14)(CO₂ + 37)]/[27.3 + (0.14)(pCO₂ + 37)]; Δ¹³C_27% = [(23.9)(0.12)(CO₂ + 47)]/[23.9 + (0.12)(pCO₂ + 47)]. Mean absolute error between the modeled and measured Δ¹³C values was 0.3‰ (θ_m = 60%), 0.4‰ (θ_m = 45%), 0.2‰ (θ_m = 30%) and 0.0‰ (θ_m = 27%).

Because both biomass and Δ¹³C value increased with increasing CO₂ and Δ¹³C value is a measure of plant carbon assimilation, we considered the effect of biomass on Δ¹³C value by examining if larger plants had higher Δ¹³C values within a given experiment. Despite the broader trend between biomass and Δ¹³C value across CO₂ levels, we found no effect of biomass on Δ¹³C value within experiments [median slope = −0.1‰ per ln(mg), p = 0.28, n = 18)], with a majority (n = 11) yielding a negative slope (i.e., lower Δ¹³C with increasing biomass). A lack of effect of biomass on Δ¹³C value within a treatment is consistent with previous results³¹, which also saw no correlation between Δ¹³C and below- or above-ground biomass (R² < 0.1), suggesting small intra-experiment variations in Δ¹³C value are inherent variability not attributable to the small variations in biomass.

Discussion

Photorespiration, the loss of fixed carbon due to successful competition of O₂ for sites on the Rubisco enzyme, is extremely responsive to CO₂ (ref. ³²). Independent observations of the glycolate pathway have confirmed that photorespiration decreases with increasing CO₂ (ref. ³³), leading to enhanced biomass³⁴ and higher Δ¹³C value³⁵.

Soil moisture constitutes an obvious control on the amount of biomass produced during growth (e.g. ref. ³⁶), thus the increase in biomass with increasing θ_m that we observed followed expected patterns, given the documented effect of soil-water deficit on the physiological processes that govern cell growth³⁷. Decades of previous work have documented an increase in biomass with increasing CO₂ level, mainly from studies examining elevated CO₂ up to only 475–750 ppm (e.g. refs. ^{38,39,40,41,42}), but also in studies that include super-elevated CO₂ levels, i.e., up to 900–1200 ppm (e.g. refs. ^{31,43,44,45,46,47,48,49,50}) and higher (i.e., >10,000 ppm) (e.g. refs. ^51,52,53). This “fertilization-effect” of elevated CO₂ on biomass was also apparent in our experiments, across all levels of soil moisture. Thus, the results of our experiments showed the clear effect of both variables on Arabidopsis biomass, in keeping with both sets of literature.

Our biomass results suggest two things: first, that the plant’s physiological experience of water-stress did indeed increase with the amount of water-stress applied in the form of gravitational water content (θ_m). Second, because we investigated plant response across multiple levels of soil moisture (θ_m = 27, 30, 45, and 60%), we were able to observe that the physiological experience of water-stress was not linearly related to θ_m value; for example, plants experienced much more physiological distress from a decrease in θ_m value from 45% to 30%, than from 60% to 45%. Both implications hold true across the wide range of CO₂ levels used in our experiments (i.e., 389–2175 ppm). The observed biomass response is a testament to the severe water stress achieved within these experiments, including lack of seedling viability at ambient CO₂ and minimum soil water content (θ_m ≤ 30%).

The observation that increasing water-stress results in decreased Δ¹³C values under a given CO₂ level has been known for nearly half a century (e.g. ref. ³⁰). We previously showed that increasing CO₂ results in increased Δ¹³C values under constant soil moisture in multiple plant-growth experiments including super-ambient CO₂ levels up to 4200 ppm (ref. ¹⁰), and sub-ambient CO₂ levels down to 97 ppm (ref. ¹⁶). It is notable that the Δ¹³C values gained from the experiments presented here, which were designed to evaluate both variables simultaneously across a wide range of water-stress, showed the same hyperbolic response as our previous studies growing plants under constant soil moisture and across a range of CO₂ (ref. ¹⁰).

Studies that report Δ¹³C values for multiple levels of CO₂ can be quantitatively compared via the slope (S) of the response. We proposed a unifying relationship between CO₂ and S-values using data from 23 published plant growth studies encompassing sub-ambient, ambient, and elevated CO₂ levels^10,54,55 (Supplementary Table 1):

$$S=({A}^{2}B)/{\left[A+B\left({{{{\rm{CO}}}}}_{2}+C\right)\right]}^{2}$$

(2)

We calculated the S-values for each of the θ_m-treatments using fitted regression data, and found them in agreement with the relationship above (Fig. 4). This means that the same increase in Δ¹³C value with increasing CO₂ occurred at all levels of soil water content, including both PWP and FC, as well as at the midpoint, θ_m = 45% (e.g. compare S-values in Table 3). Thus we expect that the change in Δ¹³C value with increasing CO₂ is the same across all conditions of soil moisture that support plant growth.

Fig. 4: The relative change in Δ13C value per unit increase in CO2 (i.e., S, ‰ ppm-1) observed here for each soil water treatment (θm = 27, 30, 45, and 60%), compared with literature values determined using fitted regression data (gray, updated from refs. ^10,54,55).

Horizontal bars encompass the range of CO₂ levels used within each experiment; the circle is plotted at the midpoint of the range. The blue, green, tan, and dark brown curves represent the first derivative of the curves shown in Fig. 3 (blue curve overlies green curve); a best-fit curve through the literature values (n = 64) is shown for comparison (gray curve with gray shading), after Eq. (2), where A = 28.3 ± 1.2, B = 0.22 ± 0.12, and C = 24 + 28/−8) with uncertainty in A and B determined as ±1σ of 8 published experiments within refs. ^10,101,102. C determined as: C = (4.4 A)/(AB − 4.4B) after ref. ¹⁰³. Inset shows CO₂ = 0–1000 ppm. Data are provided in Supplementary Table 1.

Table 3 Calculated S values (Eq. (2) and Fig. 4) for different water treatments

Conclusions

This study shows that the predictive, non-linear response between CO₂ and Δ¹³C value observed previously within only well-watered experiments persists across a wide range of soil water content. Furthermore, our results provide the experimental data to quantify the relative contribution of soil moisture and CO₂ on carbon isotope discrimination of plant tissue, and point towards two separate, but related, conclusions. First, the data show that the effect of CO₂ on Δ¹³C value is quantitatively the same across all levels of θ_m, from maximum soil hydration to the minimum soil hydration required to prevent plant death. Specifically, Δ¹³C value increased non-linearly with increasing CO₂, consistent with the effect of diminished photorespiration under increasing CO₂.

Second, the data show that the effect of soil moisture on Δ¹³C value is non-linear across all levels of CO₂. Consistent with ref. ⁵⁶. no meaningful change in Δ¹³C value was observed at FC or at intermediate levels of soil water content (i.e., θ_m = 60 and 45%); significant change in Δ¹³C was observed only when soil water content was decreased to near the permanent wilting point, suggesting that stomatal conductance (c_i/c_a) was only affected when soil moisture was severely limited.

Decrease in soil moisture is one of the many consequences of drought that affect plant carbon assimilation⁵⁷. The results presented here confirm soil moisture as a dominant control on photosynthesis, including within ecosystems subject to increasing CO₂. However, anthropogenically increasing CO₂ is actively imparting an additional, non-linear, and significant isotopic discrimination. From this we suggest that workers seeking to evaluate plant carbon assimilation under increasing dryness using stable carbon isotopes must first account for concurrently rising CO₂ within the earth’s atmosphere.

Our findings also have implications for the reconstruction of past CO₂, given that the photorespiration mechanism is a conserved trait through the billion-plus year history of photosynthesis. We have argued that if the carbon isotopic composition of the ancient atmosphere is known or assumed, Δ¹³C can be calculated from the δ¹³C records of terrestrial organic matter (TOM) and used to estimate the CO₂ of the past⁵⁴. We have used this method to reconstruct CO₂ across the last 23 Ma (ref. ⁵⁸), as well as across multiple intervals of the Paleogene^55,59,60,61. Others have also applied our work to calculate CO₂ during the middle Miocene⁶², Eocene⁶³, late Paleocene⁶⁴, Aptian (early Cretaceous)⁶⁵, early Bajocian (Middle Jurassic)⁶⁶, early Toarcian (Early Jurassic)⁶⁷, Pliensbachian–Toarcian transition (Early Jurassic)⁶⁸, Sinemurian (Early Jurassic)⁶⁹, Permian-Triassic⁷⁰, and Paleozoic⁷¹.

Despite the classically-established wetland megabias^72,73,74,75, which affirms that long-term preservation of terrestrial organic material is overwhelmingly relegated to wet-environments, a recurring critique of our model is that it does not incorporate the potential effect of water stress on Δ¹³C (refs. ^76,77), an effect that, if present, would lead to a systematic underestimate of CO₂ (ref. ⁷⁸). This study presents evidence that variations in θ_m between 45 and 60% (i.e., the soil moisture levels germane to the floodplain, wetland, and forest ecosystems) are negligible compared to the effect of CO₂ on Δ¹³C on where TOM is preserved. This implies that, short of the appearance of independent evidence for soil moisture levels decreasing to near wilting-point, changes in Δ¹³C inferred from fossilized TOM are best interpreted as a record of change in CO₂.

Materials and method

Experimental design

Several methods have been employed for the experimental application of a quantified and consistent level of soil moisture across a period of plant growth. For example, plant irrigation (e.g. ref. ⁷⁹) allows for a quantification of gross water supply (e.g., ml/day/pot), but results in heterogeneous water stress among individual plants as they deplete the soil water at different rates during growth. More intensive methods such as directly manipulating soil water potential via the addition of an osmoticum such as mannitol (e.g. ref. ⁸⁰) are not suitable for long-term growth experiments due to phytotoxicity under prolonged exposure.

For the work reported here, we built upon the gravimetric method described by ref. ⁸¹ for the maintenance of soil water content for plant growth quantified in terms of g of water per g of soil (θ_m), for use within the Plexiglas chambers that we designed to supply known and consistent levels of elevated CO₂ for up to several months of plant growth⁸². Briefly, the method from ref. ⁸¹ involves the determination of θ_m at retention capacity, followed by daily pre-dawn gravimetric augmentation to maintain a chosen level of θ_m for individual plants across several weeks of growth. An automated version of the original method (“PHENOPSIS”) was used to successfully maintain Arabidopsis growth under stabilized drought conditions until the end of rosette development (i.e., germination through the growth of 10–15 leaves)⁸³, significantly advancing the utility of the Arabidopsis genome for the study of plant water stress.

Because our goal was to measure the effects of the full range of soil moisture under which plants are viable, we performed trial experiments designed to quantify θ_m across the full range of soil water content that a plant may experience, i.e., between mortality due to excess water and mortality due to insufficient water. Our first task was to calibrate the θ_m of our experimental soil in terms of field capacity (FC) and permanent wilting point (PWP) under ambient CO₂. PWP, the minimum hydration within a soil to prevent mature plant death is classically defined as −1.5 MPa of hydraulic head. FC, the maximum hydration that an aerated soil may hold, is defined as −0.33 MPa. Because hydraulic head is strongly affected by soil porosity, values of θ_m (i.e., the variable modulated by the gravimetric method above) are specific to soil type.

To perform this calibration, we first irrigated our experimental soil (a peat-moss based potting soil containing 0.21% total nitrogen, 0.11% available phosphate and 0.16% soluble potash) to just above FC, as signified by the first appearance of standing soil water. At this level of soil moisture, trial Arabidopsis plants experienced wilting and then mortality due to root hypoxia, as expected. We determined the θ_m of this condition to be 1.63 g g⁻¹ (62%). Similarly, we observed that trial Arabidopsis plants experienced desiccation and mortality when θ_m fell below 0.44 g g⁻¹ (30%). We confirmed our experimental results against the work of ref. ⁸⁴ who measured the gravimetric water content of peat-based soils at both FC and PWP in order to characterize hydrophobicity during soil wetting and drying cycles; they determined FC = 1.50 g g⁻¹ (60%) and PWP = 0.44 g g⁻¹ (30%).

Based on the congruence of our empirical data with the published standard above, we chose θ_m = 1.50 g g⁻¹ (60%) as the high-end value for our CO₂ experiments as a representation of luxury conditions, i.e., growth unlimited by water availability. We chose θ_m = 0.40 g g⁻¹ (27%) as the low-end value even as we fully expected plant mortality under ambient CO₂. We anticipated some growth at the 27 and 30% level under elevated CO₂ as high levels of carbon dioxide have long been known to alleviate water stress via high concentration of carbon in the atmosphere offsetting reduced diffusion due to stomatal closure (e.g. ref. ⁸⁵). In addition, we included θ_m = 0.83 g g⁻¹ (45%; i.e., the midpoint between 30% and 60%) in order to gain comparative insight into relative functioning under substantial, but not terminal, water stress.

Because our ultimate aim was to produce experimental results with the potential to scale up to real-world scenarios, we followed the guidance offered by ref. ⁸⁶ which stressed the need for plant experiments to account for non-linearity in system responsiveness. By designing our experiments to include five different levels of CO₂ and four different levels of θ_m, we allowed for a higher-order analysis that lends itself to upscaling. Table 4 lists the values of θ_m, CO₂, and the number of plants analyzed (n) for each of the experiments performed.

Table 4 Soil water content (θ_m), CO₂ level, and the number of plants analyzed (n) for each of the 20 experiments performed

Plant growth

We germinated a total of 164 Arabidopsis thaliana (Col-0, Carolina Biological, USA) seeds within individual pots inside five controlled growth chambers, each maintained at one of the following CO₂ levels: 389, 685, 1076, 1583, and 2175 ppm. These levels span 200+ years of recent and projected CO₂ (years: ~2010–2250) (refs. ^87,88). CO₂ levels were maintained at their target concentrations by bleeding pure cylinder CO₂ into an intake pipe where it mixed with ambient air¹⁰, all major atmospheric gases (N₂, O₂, Ar) were not modulated from ambient conditions within our experiments, but their high concentrations (percent scale) and long mean residence times in ambient atmosphere preclude significant variation between CO₂ treatments⁸⁹, particularly given the high flow-through rate of our chambers (complete atmospheric turnover occurred once every ~10.4 min). Moreover, N₂ and Ar are highly unreactive, and previous work showed that the effect of O₂ on Δ¹³C is minor compared with changes in CO₂, even when O₂ is intentionally varied by several percent¹². We chose to test our hypotheses using common Arabidopsis because it is the most widely studied plant in modern biology and serves as the model multicellular photoautotroph for study of molecular, cellular, and developmental processes⁹⁰. This includes the first elucidation of the photorespiratory mechanism in plants⁹¹, which is subject of our experiments and the fundamental carbon isotope theory tested here.

For the period encompassing germination and early emplacement, each plant was subject to identical environmental conditions designed to support growth without limitation, in keeping with those used in our previous experiments^{10,16,31,82,92,93}. In keeping with published standards for Arabidopsis growth⁹⁴, all plants were grown in the peat-based soil described above; no additional fertilizer was applied for the duration of the experiment. Similarly, light levels (i.e., photosynthetic photon flux) were maintained at 280 µmol m⁻² s⁻¹ (400–700 nm) for 14 continuous hours each day. Because we wanted to isolate the effects of CO₂ and soil moisture only on Δ¹³C value, other variables, including average air temperature and average relative humidity were maintained constant across all treatments (average air temperature = 21.9–22.2 °C; average relative humidity = 64.0–66.1%). This was achieved using dedicated air conditioning units (Everstar model MPN1-095CR-BB6) and computer-controlled custom humidifiers⁹³.

In order to support the first eleven days of seedling growth, we manipulated the water available within each pot so that θ_m = 1.50 g g⁻¹ (60%); after this θ_m adjustment period, we either maintained this value or allowed the plant to dry to experimental conditions. Within one week, we had stabilized at least ten of the original twelve seedlings at each level of θ_m (i.e., 60%, 45%, 30%, and 27%) within the elevated CO₂ chambers, as well as in the pots with θ_m = 60% or 45% within the ambient CO₂ chamber. We allowed plants to grow until floral initiation (27–30 days after seed sowing) while individually maintaining each pot at its stabilized level of θ_m using the augmentation method described earlier. As expected from our trial results, no plants could live or grow at θ_m ≤ 30% under ambient CO₂ indicating that we have included conditions of maximum water stress in our experimental range.

Just prior to flowering, the above-ground tissues from each plant were harvested and dried at 60 °C, then weighed to assess the biomass of each plant (±0.1 mg). In order to evaluate biomass addition during the θ_m adjustment period, we grew an additional 87 plants for the specific purpose of harvesting these plants on day 17 (12 days after germination). The median biomass of these seedlings = 0.9 ± 0.4 mg, meaning that the vast majority (96–99%) of the plants’ final biomass was added after manipulation of water stress to experimental conditions.

Isotope analysis

After dried tissue was assessed for biomass, it was homogenized using a mortar and pestle in preparation for stable isotope analysis. The δ¹³C value of the above-ground tissues from each plant (δ¹³C_p) was analyzed using a Delta V Advantage Isotope Ratio Mass Spectrometer (Thermo Fisher, Bremen, Germany) coupled to a Costech ECS 4010 Elemental Analyzer with a zero-blank autosampler (Costech Analytical, Valencia, CA, USA). Data were normalized to the Vienna Pee Dee Belemnite (VPDB) scale using internal lab reference materials (JGLUT: δ¹³C = −13.43‰; JGLY: δ¹³C = −43.51‰). In order to assess the accuracy and precision of the analyses, six capsules of quality-assurance standards (JRICE) were analyzed as unknowns with each batch run. Across all analyses, the JRICE quality assurance sample averaged δ¹³C = −27.37 ± 0.05‰ (n = 60), which is in agreement with the calibrated value of −27.37‰. Each homogenized plant sample was also analyzed in triplicate; the average standard deviation of the three replicates was 0.07‰. This level of precision (1σ better than ±0.1‰) is expected for stable isotope mass spectrometry work, and commonly reported for well homogenized, easy to combust organic material (e.g., plant materials)^95,96,97.

In order to precisely calculate Δ¹³C [Δ¹³C = (δ¹³C_a − δ¹³C_p)/(1 + δ¹³C_p)], samples of air were collected from within each chamber and analyzed for the δ¹³C value of CO₂ (δ¹³C_a) using the direct injection method described previously^10,82, with calibrated reference gas δ¹³C values = −25.3‰ and −10.3‰. Precision for each reference gas across the duration of the experiment was 0.1‰ (1σ, n = 22 and 21 respectively). The standard deviation of the δ¹³C_a value within each chamber over the course of the 8 sample days did not exceed ±0.1‰ for any chamber, and thus reflected the δ¹³C_a value within each chamber throughout the course of the plant growth experiments (Fig. 5).

Fig. 5: Demonstration of stability in δ¹³C_a value measured for each level of CO₂ across the duration of the experiments.

Average δ¹³C_a value for each chamber is indicated with a dashed line and reported in Table 2. Analytical precision was 0.1‰ for each measurement. Note that δ¹³C_a values shifted downwards in chambers with higher CO₂ levels. This was expected and occurred because the cylinder CO₂ used to increase the CO₂ levels had a lower δ¹³C_a value than the ambient air with which it is mixed^10,82.

The reason we reported our data in terms of the carbon isotope discrimination by the plant [i.e., Δ¹³C, after ref. ⁹⁸], rather than δ¹³C_p, is that Δ¹³C value is independent of the source air CO₂ composition (i.e., δ¹³C_a). This is particularly important for chamber experiments in which the δ¹³C_a value differs with CO₂ treatments (lower δ¹³C_a with increasing CO₂, Fig. 5) or any environment in which δ¹³C_a is changing⁹⁹. This is the same reason that changes in plant carbon isotope value measured across recent industrialization (e.g. ref. ¹⁰⁰) are best interpreted in terms of isotopic discrimination, which removes the influence of δ¹³C_a on δ¹³C_p (ref. ⁹⁹). We note that Δ¹³C is a measure of the intrinsic isotopic discrimination by the plant, and is by definition independent of the source air composition (i.e., δ¹³C_a).

The standard deviation of our Δ¹³C data (σ_Δ) was at least half that of efforts by other researchers (i.e., ref. ⁷⁷) for both our two highest θ_m treatments (σ_Δ = 0.2‰ for θ_m = 45 and 60%) and two lowest θ_m treatments (σ_Δ = 0.8‰ and 0.7‰ for θ_m = 27% and 30%, respectively), which allowed for clear observation of the CO₂ versus Δ¹³C response across multiple levels of soil moisture. The full complement of biomass and stable carbon isotope data is included in Supplementary Table 2 and summarized in Table 2.