We used the Web of Science¹ and China National Knowledge databases (CNKI²) to search for peer-reviewed papers related to water deficit × CO₂ interactions in plants. Various keyword combinations were used for the search, including (drought OR water deficit OR water supply) AND (CO₂ enrichment OR doubled ambient CO₂ OR rising CO₂ OR CO₂ rise) AND (gas exchange OR plant physiology). A total of 554 observations from 54 published papers were included in this meta-analysis (Appendix and Supplementary Material).

Only experiments on e[CO₂] and water deficit were included, and the few available studies simultaneously testing a third variable, such as temperature or nutrient levels, were omitted. Factorial experiments included at least two water treatments in addition to two CO₂ concentration treatments. The “well-watered,” “water deficit,” “ambient [CO₂]”(a[CO₂]), and “elevated [CO₂]” (e[CO₂]) treatments followed the definition of the authors of the original experiment (Wang et al., 2015). Meanwhile, the two most common methods of water control were chosen: dry days and field water capacity (FWC). All studies included plants growing in pots. We extracted response variables for leaf gas exchange, including P_n, G_s, T_r, and instantaneous WUE, which was calculated from P_n/T_r by authors. For those studies that included only P_n and T_r data but not WUE, we supplemented the WUE values with the equation: WUE = P_n/T_r. Data were taken from tables or digitized from figures using the software GetData Graph Digitizer, 2008 (ver. 2.22, Russian Federation).

The factorial experiments included four treatments: (i) ambient CO₂ (a[CO₂]) + well-watered (C_aW_w); (ii) e[CO₂] + well-watered (C_eW_w); (iii) a[CO₂] + water deficit (C_aW_d); and (iv) e[CO₂] + water deficit (C_eW_d). In our dataset, a[CO₂] treatments ranged from 350 to 450 ppm, while e[CO₂] ranged from 500 to 1,200 ppm. The e[CO₂] treatments were grouped into seven categories: 500–550, 551–600, 601–650, 651–700, 701–750, 751–800, and >800 ppm. In addition, CO₂ exposure was either in a closed environment (growth chamber, greenhouse, and glasshouse), or an open or semi-open environment; free-air CO₂ enrichment (FACE) and open top chamber (OTC). Water treatments were categorized into well-watered and water deficit, with the water-deficit treatment grouped into three categories for FWC: 65–55, 54–40, and <40%. Dry days were not grouped because of the small amount of data. Additionally, several explanatory variables in our meta-analysis may affect leaf gas exchange under water deficit and e[CO₂], including plant type (grass, tree, legume, shrub, or crop), source of light (lamp or solar), and photosynthetic pathways (C₃ or C₄) as described in the Supplementary Tables 1, 2.

We considered a[CO₂] and well-watered treatments as the baseline, whereas e[CO₂] and water deficit were the experimental treatments. The individual effect sizes for water and CO₂ manipulation (r_w and r_c, respectively) were calculated as follows:

where X¯, C_e, C_a, W_w, and W_d represented the mean, e[CO₂], a[CO₂], well-watered, and water-deficit treatments, respectively (Jiang et al., 2020). The water deficit × CO₂ interaction term was calculated from factorial experiments as described by Lajeunesse (2011). The water deficit and e[CO₂] interaction effect size, or the effect of water deficit on the e[CO₂] responses was calculated as follows:

r was log-transformed to linearize this metric as follows:

Based on the additive property of variance (Curtis and Wang, 1998; Baig et al., 2015; Jiang et al., 2020), the variance of the water deficit by the CO₂ interaction (v) response ratio was calculated as follows:

An overall interaction term was estimated using weighted means, with greater weighting given to experiments with greater precision. A random-effects model was selected because the between-study variance was statistically significant. A multivariate linear mixed-effects model was then used to estimate the mean and the 95% confidence interval (CI) of the log-transformed response ratios for each individual variable, weighted by the variance of individual studies described in Supplementary Tables 1, 2. Effect sizes were reported as the antilog r converted to the mean percentage change from the baseline treatment [(r − 1) × 100]. Meta-analysis and Ln(r) calculations were conducted using OpenMEE software for ecological and evolutionary meta-analysis (Wallace et al., 2017).

Graphical vector analysis was first developed to simultaneously compare the effect of experimental treatments on plant biomass, nutrient concentration, and nutrient content using an integrated graph (Haase and Rose, 1995). It is widely used because the comparison may be interpreted independently of predetermined critical levels or ratios.

In this analysis, GVA was used according to Wang et al. (2015); Couture et al. (2017), and Li Z. et al. (2020) to complement the meta-analysis of WUE. The magnitude of change in leaf-level WUE is determined by P_n and T_r. Responses of leaf-level WUE to the experimental treatment relative to the control fell into one of the following seven categories:

All variables were plotted to illustrate the percent change in water deficit, e[CO₂], and their interactions (value at C_aW_d divided by those at C_aW_w and multiplied by 100, value at C_eW_w divided by those at C_aW_w and multiplied by 100, value at C_eW_d divided by those at C_aW_w and multiplied by 100). T_r (y-axis) was plotted against P_n (x-axis) with WUE as the diagonal axis (y-axis) in a square-shaped diagram. Plotting the P_n data and WUE data automatically positioned the points along the diagonal lines representing the z-axis value for T_r, allowing simultaneous examination of the three related variables on a two-dimensional diagram.

The dataset was derived from 54 studies of 554 valid data points of gas exchange (P_n, G_s, T_r, and WUE) under e[CO₂] and water deficit. Under e[CO₂], P_n was increased by an average of 11.9 and 16.4% for well-watered and water-deficit treatments, respectively, for all plants, whereas the interaction of water deficit and e[CO₂] slightly decreased P_n by 8.3%. That is, the e[CO₂]-induced stimulation of P_n was reduced by the water deficit (Figures 1B,C). In contrast, water deficit reduced P_n by 40.5 and 37.8% because of a[CO₂] and e[CO₂] treatments, respectively, and WUE was increased by 24.5 and 21.5% (Figure 1B). The negative effects of water deficit on plant P_n were relieved at e[CO₂] (Figures 1A–C).

In contrast to P_n, e[CO₂] reduced G_s and T_r by 23.3 and 14.4% under water-deficit treatments, respectively. However, a greater e[CO₂] reduction was observed in well-watered plants (28.5 and 19.1%, respectively). Similarly, water-deficit treatment reduced G_s and T_r by 56.7 and 45.2%, respectively, under e[CO₂] treatment (Figures 1A,B). In comparison, WUE was increased due to the e[CO₂] and water deficit. e[CO₂] increased WUE by 29.3 and 28.8% under well-watered and water-deficit treatments, respectively, and water deficit increased WUE by 24.5 and 21.5%, respectively, under a[CO₂] and e[CO₂] treatment (Figures 1A,B). In addition, the interaction between water deficit and e[CO₂] and the impact on G_s, T_r, and WUE were not statistically significant.

P_n responded to changes in water and CO₂ treatments, types of light sources, and fumigation methods. The stimulation of P_n by e[CO₂] was 50.7% lower under water-deficit treatments than the well-watered treatments using solar radiation (Figure 1), whereas the stimulation of CO₂ on P_n under water deficit was almost the same as that under well-watered treatment using lamp radiation. Additionally, under e[CO₂] conditions, water deficit decreased plant P_n by 41.8 and 35.6% when using the field capacity and drought days methods to control the water conditions, respectively. A greater increase in P_n was observed when plants were exposed to e[CO₂] in a free-air CO₂ enrichment system (FACE, 64.2%) than in closed glasshouses (35.3%), greenhouses (14.1%), or growth chambers (4.3%) (Figure 1B). However, the e[CO₂]-induced increase in P_n was significantly reduced by water deficit compared with well-watered plants in any growing environment (Figure 1C).

The stimulatory effect of increasing [CO₂] on P_n reached a maximum at 551 to 600 ppm in well-watered plants. However, the e[CO₂]-induced increase in P_n was significantly reduced by water deficit (43.4%) at 551–600 ppm (Figure 2A). Very strong CO₂ fertilization effects on P_n were observed when the [CO₂] increased from 601 to 800 ppm, and became >800 ppm during water deficit compared with the well-watered treatment (Figure 2A). The water deficit effect size magnitudes for P_n were −0.343, −0.408, and −0.75 at 55–65%, 40–54%, and <40% soil water at a[CO₂], respectively, indicating that the highest water deficit resulted in the greatest inhibition (Figure 3A).

Plant P_n significantly increased by 11.2 and 21.5% when the soil water was between 65–55% and 54–40%, respectively, in conditions of e[CO₂] compared to a[CO₂] treatment. However, e[CO₂] increased the inhibition after water-deficit treatment by 3.9% when the soil water was <40% (Figure 3A). The effect size of P_n decreased linearly with increasing drought days, with R² values of 0.19 (Figure 4A; p = 0.0321) and 0.26 (Figure 4A; p = 0.0164) at a[CO₂] and e[CO₂], respectively, in a closed growing environment. However, there was no significant correlation between plant P_n response and drought days in open and semi-open environments, possibly owing to the relatively small sample size (n = 8) (Figure 4B).

Water-deficit treatment reduced the plant G_s for most plants except in shrubs, regardless of the CO₂ treatment. Tree and crop G_s were significantly reduced by e[CO₂] after water-deficit treatment by 11.1% (n = 53) and 14.5% (n = 20), respectively. Additionally, the water-deficit effect sizes of grass and legume G_s were −1.067 and −1.344 under a[CO₂], which increased by 16.3 and 0.6% compared with that in e[CO₂] conditions, respectively, (Figure 1). Similarly, e[CO₂] reduced all plant G_s, regardless of the drought treatment, except for shrubs. The e[CO₂] effect sizes of crop, legume, tree, and grass G_s were −1.229, −1.718, −1.057, and −1.519 under well-watered conditions; and −0.406, −0.376, −0.251, and −0.204 under water-deficit treatments, respectively (Figure 1). Therefore, water deficit reduced the e[CO₂] response. In contrast, the effect size of shrubs was positive (0.135); however, the 95% CI included zero. The effects of the two water control methods on G_s were also not significant. The interactions of water deficit and e[CO₂] on G_s were not statistically significant for plant type and control methods of water because the 95% CI included zero.

When the CO₂ concentration was higher than 650 ppm (651–700, 701–750, 751–800, and >800 ppm), G_s decreases significantly in conditions of well-watered treatment (Figure 2B). Similarly, water-deficit treatment reduced G_s. The water-deficit effect size of G_s was −0.577, −0.74, and −1.594 when the FWC was 65–55, 54–40, and <40% after a[CO₂] treatment, respectively. However, the water-deficit effect size of G_s was −0.552, −0.487, and −1.226 when the FWC was 65–55, 54–40, and <40% after e[CO₂] treatment (Figure 3B).

Water deficit reduced the T_r of C₃ plants to a greater degree than C₄ plants, whereas e[CO₂] reduced the T_r of C₄ plants more than that of C₃ plants. Meanwhile, water deficit reduced the T_r of grasses, trees, and legumes more than crops regardless CO₂ concentration. Additionally, water deficit reduced the T_r by 29.7, 26, 0.52, and 56.5% in OTC, greenhouse, FACE, and growth chambers, respectively, under a[CO₂] (Figure 1B). However, water deficit increased the T_r by 9.6% in FACE under e[CO₂] treatment (Figure 1A). Moreover, e[CO₂] reduced T_r regardless of the water condition and CO₂ fumigation method except for OTC. However, the interactions of drought and e[CO₂] on T_r were statistically insignificant for legumes and crop species.

The decrease in T_r for plants resulted in a much larger decrease in T_r at >750 ppm CO₂ compared with 551–750 ppm CO₂ (Figure 2C). However, there was no significant influence on T_r at 500–600 ppm CO₂ (Figure 2C). The water deficit effect size of T_r was −0.436, −0.68, and −1.306 when FWC was 65–55, 54–40, and <40%, respectively (Figure 3C).

Graphical vector analysis showed that increased leaf-level WUE under C_eW_w treatment compared to the control (C_aW_w) may be due to the common effect of P_n and T_r (Figure 5). However, increased WUE caused by water deficit (C_aW_d) is attributable to the “T_r” positive effect (Figure 5). Water deficit induced a larger increase in WUE under solar grown plants compared to plants grown under a lamp, regardless of [CO₂]. e[CO₂]-induced stimulation of WUE was attributable to a “P_n” positive effect when grown in sunlight (Figures 6A,B). There were no significant interaction effects for C₃ plants in WUE (Figures 1A–C). Water deficit increased the WUE of C₃ plants by 32.8 and 21.4% in the a[CO₂] and e[CO₂] treatments, respectively, but it had no effect on the WUE of C₄ plants. Instead, water deficit resulted in a parallel decrease in P_n and T_r of C₄ plants (“steady-state” decrease) under a[CO₂] (Figures 6C,D). Meanwhile, e[CO₂] increased the WUE of C₃ and C₄ plants by 31.7 and 53.3%, respectively, in the well-watered treatments, and by 20 and 95.8%, respectively, in the water-deficit treatments.

Water use efficiency reached a maximum of 0.593 (81.0%) and 0.483 (62.1%) when the CO₂ concentration was 601–650 ppm in water deficit and well-watered treatments, respectively. The e[CO₂]-induced increases in WUE were less prominent in the water deficit treatment than in the well-watered treatment at 500–550, 551–600, 601–650, and 651–700 ppm [CO₂] (0.246 vs. −0.05; 0.334 vs. 0.18; 0.593 vs. 0.483; 0.31 vs. 0.228; and 0.512 vs. 0.278, respectively) (Figure 2D). However, the e[CO₂]-induced increase in WUE during water deficit treatment was 1.8 and 2.53 times greater than that of the well-watered treatments at 751–800 ppm CO₂ and >800 ppm CO₂, respectively (Figure 2D). The water deficit induced increase in WUE in the e[CO₂] treatment was 47.1% lower than that in the a[CO₂] treatment when FWC was 54–40%, whereas the increase in WUE in the e[CO₂] treatment was 78.8% higher than that in the a[CO₂] treatment when FWC was <40% (Figure 3D). Additionally, there was no significant change in WUE when the FWC was between 55 and 65% (Figure 3D).

It has been suggested that the positive effects of e[CO₂] cannot be maintained when other environmental factors (nutrients, temperature, and availability of water) are limited (Carter et al., 2007; Elliott et al., 2014; Fernando et al., 2019). Our meta-analysis results supported our H1 hypothesis that the e[CO₂]-induced stimulation of P_n was reduced by water deficit treatment (Figures 1B,C). e[CO₂] leads to a delay in the onset of drought stress due to increased stomatal closure preventing water loss (Paoletti and Grulke, 2005; Yamaguchi et al., 2019; Jiménez et al., 2020). However, this delay only existed under mild drought stress because further stomatal closure leads to decreases in intercellular CO₂ and reduces carbon assimilation (Figure 7). e[CO₂] did not weaken the inhibition by water deficit treatment when soil water was <40% FWC (Figure 3A). This result indicates that the effect of soil water deficit is stronger than that of e[CO₂] when under severe water-deficit stress. In general, water deficit negatively affected the P_n of most C₃ plants due to favoring RuBP oxygenation over carboxylation. Meanwhile, water deficit induced the stomatal closure reduced T_r, thereby inhibiting the uptake of N, which was associated with reductions in the amount and/or activity of Rubisco (Figure 7). Additionally, Zheng et al. (2020) found that the regularity of stomatal distribution pattern was dramatically reduced by e[CO₂] when winter wheat plants were constrained to moderate and severe stresses, implying that soil moisture conditions partly determined the response of stomatal distribution pattern to e[CO₂] (Figure 7). The effect size of P_n decreased linearly with increasing drought days, regardless of [CO₂], also suggesting that water stress had a greater impact on plant growth than e[CO₂] when plants are under severe stress.

Stomatal optimization theory states that stomatal opening to allow CO₂ uptake inevitably comes at the expense of H₂O loss (Ferris et al., 2002; Robredo et al., 2010). Therefore, climate change will affect not only the rate of carbon fixation in plants, but also water loss. Evidence was found to support our H2 hypothesis that both water deficit and e[CO₂] treatment improved leaf-level WUE, but the underlying mechanisms for maintaining high water availability may be different. In general, an increased WUE is a rapid response to drought by closing stomata through root-generated xylem-born ABA (Ferris et al., 2002; Zhang et al., 2018, 2021). Conversely, e[CO₂] changes stomatal aperture (SA) by increasing intercellular [CO₂] for a short-term response (Xu et al., 2016; Li S. et al., 2020). Additionally, stomatal density (SD) also decreased when plants were exposed to high [CO₂] for a long time (Lin et al., 2001; Zhang et al., 2019). Concurrently, an increase on P_n, driven by the improve in increase by improved regularity of stomatal spatial distribution (Xu, 2015; Zheng et al., 2020). This study showed that water deficit and e[CO₂] significantly increased WUE by 24.5 and 29.3%, respectively (Figure 1A). Increases in WUE under the C_aW_d treatment are attributable to the “T_r” positive effect (Figure 5), whereas increases in WUE under the C_eW_w treatment may be due to the common effect of P_n and T_r (Figure 7). Meanwhile, water deficit increased the WUE of C₃ plants by 32.8 and 21.4% in the a[CO₂] and e[CO₂] treatments, respectively, implying that the water deficit induced increase of WUE may also be modulated by [CO₂]. In contrast, e[CO₂] had a greater effect on WUE of C₄ plants than C₃ plants in water deficit treatment, which is consistent with previous studies (Peaarcy and Ehleringer, 1984). C₄ plants have higher light energy use efficiency because photosynthesis is completed by the combination of vascular bundle sheath cells and mesophyll cells (Peaarcy and Ehleringer, 1984; Ainsworth and Long, 2005). Therefore, C₄ plants have higher CO₂ assimilation rates, less transpiration and water loss, and higher WUE compared with C₃ plants (Peaarcy and Ehleringer, 1984; Conley et al., 2001). In addition, e[CO₂]-induced stimulation of WUE is attributable to a “P_n” positive effect when growing under sunlight regardless of water concentration. Hence, higher light appears to stimulate the effect of e[CO₂] on plants more efficiently than lamps.

It is widely reported that e[CO₂] stimulates P_n through the “CO₂ fertilization effect” because the current CO₂ concentration limits the photosynthetic ability of plants (Xu, 2015; Fitzgerald et al., 2016). Our meta-analysis demonstrated that the % increase in P_n by e[CO₂] treatments depended on the fumigation method; P_n was stimulated to a greater magnitude when plants were exposed to e[CO₂] in an open environment than in a closed environment, especially when compared to the growth chamber. In contrast, another meta-analysis studying the effects of e[CO₂] indicated that the more closely the fumigation conditions mimicked field conditions, the smaller the stimulation of rice yield (Ainsworth, 2008). It is important to note that earlier studies were conducted at lower control [CO₂] than recent studies owing to changes in atmospheric CO₂. For example, earlier FACE and chamber studies were conducted at 339 and 330–360 ppm control CO₂, respectively (Baker et al., 1990; Teramura et al., 1990), whereas recent control experiments used approximately 400 ppm (Zinta et al., 2014; van der Kooi et al., 2016; Zheng et al., 2020). In addition, there were differences in the light conditions used. For example, FACE and OTC rely on natural light, and some greenhouse/glasshouse studies also used natural light but were covered by transparent material at the top, whereas almost all plants growing in growth chambers used lamps as a light source. Enclosed growing environments may be limited through the downregulation of photosynthesis (Ainsworth and Long, 2005; Zhang et al., 2021). Under high light conditions, plants are generally better able to take advantage of increased [CO₂] due to altered N partitioning within the photosynthetic apparatus to favor light-harvesting complexes (Yu et al., 2012; Kizildeniz et al., 2021). However, at low light levels, light-dependent reactions limit the rate of photosynthesis (Assmann, 1999; Yu et al., 2012). Our meta-analysis supported these conclusions since the P_n of plants grown under natural light was higher than that under lamps. Therefore, light levels and CO₂ concentration may contribute to a smaller stimulation of P_n by e[CO₂] in a closed–CO₂ fumigation system.

Our results showed that e[CO₂] reduced G_s, but the extent of the decrease varied according to plant type. The G_s of legumes are more sensitive to [CO₂] which may be due to the decrease in total K content of legumes in high CO₂ environments, and the decrease in K⁺ concentration increases stomatal resistance resulting in stomatal closure (Peaarcy and Ehleringer, 1984). Moreover, G_s reached a minimum when the [CO₂] was >800 ppm, while the P_n did not decrease, implying that G_s was not the cause of P_n variation (Figure 1B). However, this result contradicts previous conclusions stating that changing photosynthesis and G_s is evidence of stomatal control over photosynthesis (Lawson and Blatt, 2014; Xu et al., 2016). It is possible that the decreased G_s may be the result rather than the cause of decreased photosynthesis. This may be related to e[CO₂], resulting in an increase in the concentration of ions and organic molecules; thus, guard cells swell to balance the water potential inside and outside the cell, reducing stomatal openness. Nevertheless, it should be noted that this study did not compare the long-term and short-term fumigation time of plants with high CO₂ concentration mainly due to the limited data and small sample size in previous literatures. Therefore, further studies with long-term multi-factor experiments are needed to fully understand the mechanisms and processes governing the interactions between e[CO₂] and water deficit on many plant types for comparing the “long-term” and “short-term” effects of e[CO₂] on plants under future climate change.