Elevated CO2 alleviates adverse effects of drought on plant water relations and photosynthesis: A global meta‐analysis

The elevated CO2 concentration (eCO2) is expected to improve plant water relations and carbon (C) uptakes, with a potential to mitigate drought stress. However, the interactive effects of eCO2 and drought on plant physiology and growth are not clear. We performed a meta‐analysis on the interactive effects of eCO2 and drought on plant water relations, photosynthesis, biomass production and allocation. We found that eCO2 did not lead to the conservation of soil water, but improved leaf water status under drought conditions as evidenced by a higher leaf relative water content (LRWC) and a less negative midday leaf water potential, resulting from reduced stomatal conductance (gs) and increased root to shoot ratio. Elevated CO2 retarded the response of gs to drought, which may be mediated by the decrease in leaf abscisic acid concentration under eCO2 and drought. Drought imposed stomatal limitations on photosynthesis (A), which was alleviated by eCO2 via increasing intercellular CO2 concentration (Ci). This led to a stronger A response to eCO2 under drought, supporting the ‘low Ci effect’. However, no interaction of eCO2 and drought was detected on plant biomass production. Intrinsic water use efficiency (iWUE) increased proportionally with eCO2, while plant‐scale WUE was less responsive to eCO2. C3 plants had advantages over C4 plants in terms of A and biomass production under eCO2 and well‐watered conditions rather than under eCO2 and drought conditions. Drought caused a greater reduction in biomass for woody plants than for herbs. Plants growing in pots showed greater decreases in the physiology and biomass under drought than those growing in field. Synthesis. These findings suggest that eCO2 can alleviate the adverse impacts of drought on plant water relations and C sequestration, and are of significance in the prediction of plant growth and ecosystem productivity under global changes.


| INTRODUC TI ON
The concentration of atmospheric carbon dioxide (CO 2 ), which accounts for approximately 80% of the greenhouse trace gases, plays an important role in global climate regulation (Lashof & Ahuja, 1990).
Elevated CO 2 (eCO 2 ) would inevitably cause climate warming, inducing more frequent and intense drought events (Dai, 2013;Spinoni et al., 2020). Elevated CO 2 and drought stress interactively affect plant physiology and growth in different ways, but the combined effects are far from clear (Becklin et al., 2017;Pan et al., 2022), which introduce uncertainty in the assessment of plant responses in future global change scenarios.
Drought stress has various effects on plant physiology. One immediate response of plants to drought stress is the reduction in stomatal conductance (g s ) to prevent water loss (Buckley, 2019). It is well known that stomatal closure during drought is a consequence of negative feedbacks such as hydraulic and chemical signalling [e.g. abscisic acid (ABA); Buckley, 2019;Flexas & Medrano, 2002]. The decrease in CO 2 availability at the level of the chloroplast due to the smaller g s under drought would inevitably inhibit photosynthesis (A). In addition, there are non-stomatal limitations that restrict A in the face of water deficit (Flexas & Medrano, 2002). The droughtinduced decrease in A may cause carbon (C) limitation to plant growth (McDowell et al., 2008). In contrast, it is suggested that plant growth and A are decoupled in water-limited environments because plant growth is sink-limited under drought (Muller et al., 2011).
On the other hand, eCO 2 may interact with drought stress through the 'water saving effect' and/or the 'low intercellular CO 2 concentration (C i ) effect'. The 'water saving effect' depicts that a lower g s under eCO 2 reduces plant transpiration, resulting in a higher soil water content (SWC; , which has been observed in both grasslands and forests (Leuzinger & Körner, 2007;Morgan et al., 2004). For example, Niklaus et al. (1998) showed that eCO 2 increased SWC, which would delay the onset of drought stress. However, Paudel et al. (2018) showed that SWC under eCO 2 was typically higher than that under ambient CO 2 concentration (aCO 2 ), but the difference diminished when exposed to drought.
There is also evidence that eCO 2 slowed down the rate of soil drying in the shorter term but not in the longer term . These discrepancies may suggest that the net effect of eCO 2 on SWC depends on the relative importance of the reduction in g s and the increases in leaf area and leaf temperatures (Gray et al., 2016;Jiang et al., 2021). A meta-analysis is needed to synthesize whether and under what conditions a positive effect of eCO 2 on SWC occurs.
Another potential mechanism underlying the interaction between eCO 2 and drought is the 'low C i effect' Kelly et al., 2016). It states that the drought-induced reduction in C i makes A operate on the steep initial linear phase of the A-CO 2 curve , and thus the relative response of A to eCO 2 would become more pronounced under water-limited conditions Idso & Idso, 1994;Kelly et al., 2016). The enhanced A under eCO 2 , together with the potential 'water saving effect', is expected to amplify the relative response of biomass to eCO 2 under drought conditions Kelly et al., 2016); but the experimental evidence for this expectation is equivocal, with some experiments for it (Morgan et al., 2004;Ottman et al., 2001) while others against it (Gray et al., 2016;Kelly et al., 2016).
By reducing g s and increasing A, eCO 2 induces a proportional increase in the intrinsic water use efficiency (iWUE; Medlyn et al., 2011;Walker et al., 2021;Wang & Wang, 2021). Evidence from tree rings also suggests that iWUE increased with eCO 2 (van der Sleen et al., 2015). The higher iWUE under eCO 2 slows the rate of soil moisture depletion (Peñuelas et al., 2011), which may enhance the ability of plant drought resistance (Blum, 2009). However, it is unclear whether such phenomena would maintain under drought conditions. For example, an intercomparison study (De Kauwe et al., 2013) showed that ecosystem models disagreed with one another in how drought affected the proportional relationship between iWUE and eCO 2 . In addition, it is also controversial whether the eCO 2 -induced enhancement in iWUE is scale dependent. For example, Barton et al. (2012) reported that the WUE of a Eucalyptus saligna forest was enhanced equally at both leaf scale and canopy scale, whereas Kelly et al. (2016) showed that the whole-plant WUE of Eucalyptus seedlings was less responsive to eCO 2 than the leafscale WUE. Therefore, it is needed to determine whether drought would modulate the response of iWUE to eCO 2 and how the sensitivity of WUE would change from the leaf level to the plant level.
Decades of experiments manipulating CO 2 and water availability provide evidence that the magnitude and direction of plant responses may depend on plant functional groups as well as experimental factors. Plants with different photosynthetic pathways are demonstrated to have different responses of A to eCO 2 under ample soil-water supply (Hasegawa et al., 2018;Leakey et al., 2006;Leakey et al., 2009). Specifically, eCO 2 stimulates the A of C 3 plants regardless of water availability, whereas it does not stimulate the A of C 4 plants until the onset of drought stress. This discrepancy is because C 3 plants and C 4 plants have different mechanisms to concentrate CO 2 (Wand et al., 1999). Similarly, previous meta-analyses have shown that woody plants are more responsive to eCO 2 than herbaceous plants (Ainsworth & Long, 2005;Ainsworth & Rogers, 2007).
Recently, Pan et al. (2022) reported that the above-ground net primary productivity of woody systems showed a stronger enhancement than that of grasslands under eCO 2 .
With regard to experimental factors, whether plants are grown in pots or in field may affect A response to eCO 2 , because pot size may restrict root sink strength, leading to a photosynthetic acclimation (Arp, 1991). Experimental duration also affects growth responses to eCO 2 probably due to declining nitrogen (N) availability as experiments progress (Norby et al., 2010). have been demonstrated to influence the above-ground biomass response to eCO 2 (de Graaff et al., 2006). Furthermore, drought treatments (e.g. withholding watering, imposing drying-rewetting cycles and keeping a constantly lower SWC) may affect plant responses to drought and their interaction with eCO 2 (He & Dijkstra, 2014).
Generalizing the patterns associated the variation in plant responses and testing whether they can be explained by functional groups and/or experimental factors can enhance the predictive power in global change scenarios (Ainsworth & Long, 2005;Ainsworth & Rogers, 2007).
To comprehensively assess the interactive effects of eCO 2 and drought on plant physiology and growth, we performed a metaanalysis based on 226 papers published from 1983 to 2022 worldwide to evaluate the effects of eCO 2 , drought and their interaction on plant water status, photosynthesis, WUE, biomass production and allocation. We hypothesized that eCO 2 would result in higher SWC due to its negative effect on g s (H1); drought would reduce A due to stomatal limitation, whereas eCO 2 would stimulate A by increasing C i , leading to a greater increase in A in drought treatment than in well-watered treatment (H2); iWUE would increase in response to eCO 2 , which would not differ between different watering treatments, but it would decrease when scaled to the plant level (H3); the responses of C 4 plants to eCO 2 would be more strongly modulated by drought than those of C 3 plants (H4); woody plants would show a greater response to the drought × eCO 2 interaction than herbs (H5); and pot experiments would be more restricted by drought, and less responsive to eCO 2 than field experiments (H6).

| Literature searching and data compiling
We searched the Web of Science and China National Knowledge Infrastructure using the following key words: "elevated CO 2 " or "CO 2 enrichment" or "increasing CO 2 " + "drought" or "water stress" or "reduced precipitation" + "plant responses". The studies had to meet the following criteria for selection: (1) Factorial experiments had four treatments: aCO 2 and well-watered treatment; aCO 2 and drought treatment; eCO 2 and well-watered treatment; eCO 2 and drought treatment. (2) Experimental and control plots were established within the same site, that is, same microclimate, vegetation and soil among the treatments. And (3) observations that received other treatments (e.g. warming) were excluded ( Figure S1). In total, 226 papers worldwide published from 1983 to 2022 met the criteria and were included in this synthesis ( Figure S2; Reference S1). The Africa (2 studies; Figure S2).
The response variables extracted included the following: leaf relative water content (LRWC, %), predawn leaf water potential (Ψ predawn , MPa), midday leaf water potential (Ψ midday , MPa), WUE at the plant level (WUE plant , mg/g and g/L), SWC (%), stomatal conductance (g s , mol H 2 O m −2 s −1 ), photosynthesis (A, μmol CO 2 m −2 s −1 ), intrinsic WUE [iWUE = A/g s , μmol CO 2 (mol H 2 O) −1 ], intercellular CO 2 concentration (C i , μmol mol −1 ), ratio of C i to atmospheric CO 2 concentration (C i :C a ), leaf ABA concentration (mg/g), above-ground biomass (g), below-ground biomass (g), total plant biomass (g), root to shoot ratio (R/S) and leaf area (cm 2 ). Above-ground biomass and below-ground biomass were directly obtained from the original papers or derived from R/S and plant biomass, and vice versa. Notably, to maximize the power of this meta-analysis, we included both volumetric and gravimetric SWC from both field and pot experiments.
The mean and standard deviation (SD) of each treatment were extracted from the tables or figures of the original papers using GetData Graph Digitizer 2.26. If a mean and a standard error (SE) were given, the SD was calculated as: where n is the sample size. If a mean and a confidence interval (CI) were available, the SD was calculated as: where CI u and CI l are the upper and lower limits of 95% CI, respectively; and Z α/2 is the Z score at α = 0.05. In the cases that there were no SE, SD or CI, SDs were assigned as 1/10 of the means (Luo et al., 2006).
The information on species and experimental factors was also collected wherever possible. If the response variables were reported over time, only the observations over the longest treatment duration were collected. However, the last points of some variables were almost zero in some studies; in these cases, the points before the zero points were used. For studies where same plant species had multiple eCO 2 levels and/or drought intensities, we considered them as separate observations. Therefore, the dataset contained some repeated data entries from the same study, and corresponding multiple eCO 2 or drought treatments with the same aCO 2 or well-watered treatments. The non-independent observations were tackled using the 'shifting the unit of analysis' approach (Cheung, 2015;Liang et al., 2020) in Section 2.3. In our database, species were categorized by photosynthetic pathways (C 3 plant and C 4 plant, C 3 herb and C 4 herb, C 3 grass and C 4 grass, and C 3 crop and C 4 crop) and plant growth forms (woody plant and C 3 herb, tree and shrub, and C 3 grass and C 3 forb).
We also assessed the effects of experimental protocol (e.g. (1) Experiments using growth chamber, greenhouse and glasshouse were all lumped into growth chamber. Drought manipulation was grouped into three types: keeping a constant SWC throughout the experimental duration (Type I), undergoing drying-rewetting cycles (Type II) and withholding water supply and allowing SWC decreasing over time (Type III).

| Meta-analysis
The aCO 2 and well-watered treatments were considered as the baselines for the eCO 2 and drought treatments, respectively. The respective response ratios to CO 2 and water manipulation (r c and r w , respectively) were calculated as: where X represents the mean, C e and C a represents eCO 2 and aCO 2 treatments, and W and D represents well-watered and drought treatments, respectively. Following Jiang et al. (2020), the interactive response ratio to eCO 2 and drought (r, i.e. the interactive effect of drought and eCO 2 ) was calculated as: which was linearized as: This interaction term is equivalent to the difference between the log eCO 2 response ratio at drought treatment and the log eCO 2 response ratio at well-watered treatment. The variance of r (v) was calculated as: where n C e D , n C e W , n C a D , and n C a W are the sample sizes of eCO 2 and drought treatment, eCO 2 and well-watered treatment, aCO 2 and drought treatment, and aCO 2 and well-watered treatment, respectively.
To check how normalizing plant responses to eCO 2 with the magnitude of the CO 2 treatments influences the results of the response ratio, we also calculated a β-factor for each response variable following previous meta-analyses (Walker et al., 2021;Wang & Wang, 2021). Specifically, the β-factor was calculated as: where X t and X c are means of a concerned variable in the treatment and control groups, respectively. The variance of the β-factor (v ) was calculated as: Comparing the results of the β-factor and those of the response ratio (Figures S13-S17; Table S13), we found that the β-factors of eCO 2 for almost all the 16 response variables at both well-watered and drought treatments were consistently 38.8%-42.0% greater than the corresponding response ratios (Table S13), except for the R/S ratio and ABA concentration at well-watered treatments (they were not significantly affected by eCO 2 ). Considering (1) the normalization of plant responses to drought was impossible because the degree of drought treatments in some studies cannot be determined; (2) the β-factors and response ratios had the same direction but with relatively consistent differences in the magnitude, which maintained the conclusions; and (3) a clarity of presentation, we only reported the response ratios in the main text, and kept the β-factors in the supporting information.

| Independence and weights
The 'shifting the unit of analysis' approach (Cheung, 2015) was used to tackle the non-independent observations described above. The initial weight (w) of each observation was calculated as: The weight of non-independent r (w') was adjusted by the total number (n′) of a given variable of the same species from the same study (Liang et al., 2020): A random effect model was applied to estimate the mean and the 95% CI of the log-transformed response ratios for each variable, which were weighted by the variance of individual studies.
Significant responses were recognized if the 95% CI did not overlap with zero. The between-group heterogeneity was compared by the omnibus test, with the effects of moderators considered being significant for p < 0.05. The meta-analysis was conducted with the Metafor function in r package.
We checked possible publication bias and data quality using the funnel plots and leave-one-out function. The variables were (4) for eCO 2 effect in drought treatment: r D c = X C e D X C a D ; (5) for drought effect in aCO 2 treatment: r (6) for drought effect in eCO 2 treatment: r largely independent of the influence of publication bias and outliers ( Figures S3 and S4).

F I G U R E 1
Effects of elevated CO 2 (eCO 2 ) and drought (D) on plant water relations. (a) The effect of D at ambient CO 2 (aCO 2 ) and eCO 2 . (b) The effect of eCO 2 at well-watered treatment (W) and D. (c) The interactive effect of eCO 2 and D. Response variables are: leaf relative water content (LRWC), predawn leaf water potential (Ψ predawn ), midday leaf water potential (Ψ midday ), water-use efficiency at the plant level (WUE plant ), and soil water content (SWC). The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The numbers on the right represent the numbers of observations included

| The effect of C 3 and C 4 photosynthetic pathway on plant responses
The positive effect of eCO 2 on Ψ midday was stronger for C 4 plants than for C 3 plants at the drought treatment (p = 0.05; Figure 4b; Table S1). The eCO 2 -induced increase in WUE plant in C 4 plants was significantly smaller than that in C 3 plants under well-watered conditions (p < 0.001; Figure 4a), but was similar to under drought conditions (p > 0.05; Figure 4b); drought treatment had a positive effect on the WUE plant response to eCO 2 for C 4 plants instead of C 3 plants (p = 0.01; Figure 4c; Table S1). Under well-watered conditions, the effects of eCO 2 on A (p = 0.005), above-ground biomass (p < 0.001), below-ground biomass (p = 0.03), total plant biomass (p = 0.02) and leaf area (p = 0.01) were positive for C 3 plants but not for C 4 plants (Figure 4a), whereas under drought conditions, comparable positive effects of eCO 2 were observed for C 3 plants and C 4 plants (p > 0.05; Figure 4b); the interactions between eCO 2 and drought on A (p = 0.04), above-ground biomass (p = 0.006) and leaf area (p = 0.05) were positive for C 4 plants rather than for C 3 plants, with the difference being significant (Figure 4c; Table S1). Additionally, the responses of C 3 herbs and C 4 herbs ( Figure S5; Table S2), C 3 grass and C 4 grass ( Figure S6; Table S3) and C 3 crop and C 4 crop ( Figure S7; Table S4) to eCO 2 and drought generally mirrored those of C 3 plants and C 4 plants.

| The effect of plant growth form on plant responses
Within C 3 functional groups, plants showed different responses to drought and eCO 2 . Decreases in Ψ midday caused by drought were greater for herbs than for woody plants at both aCO 2 (p = 0.04; Figure 5a; Table S5) and eCO 2 treatments (p = 0.02; Figure 5b; Table S5). The drought-induced reductions in g s (p = 0.04) and C i (p = 0.04) were significantly greater for herbs than for woody plants F I G U R E 2 Effects of elevated CO 2 (eCO 2 ) and drought (D) on plant leaf gas exchanges. (a) The effect of D at ambient CO 2 (aCO 2 ) and eCO 2 . (b) The effect of eCO 2 at well-watered (W) and D. (c) The interactive effect of eCO 2 and D. Response variables are: stomatal conductance (g s ), photosynthesis (A), intrinsic water-use efficiency (iWUE = A/g s ), intercellular CO 2 concentration (C i ), the ratio of C i to atmospheric CO 2 concentration (C i :C a ) and leaf abscisic acid concentration (ABA). The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The numbers on the right represent the numbers of observations included F I G U R E 3 Effects of elevated CO 2 (eCO 2 ) and drought (D) on plant biomass production and allocation. (a) The effect of D at ambient CO 2 (aCO 2 ) and eCO 2 . (b) The effect of eCO 2 at well-watered treatment (W) and D. (c) The interactive effect of eCO 2 and D. Response variables are: above-ground biomass (AGB), belowground biomass (BGB), total plant biomass (TB), root to shoot ratio (R/S) and leaf area (LA). The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The numbers on the right represent the numbers of observations included at eCO 2 (Figure 5b; Table S5). The decreases in above-ground biomass, below-ground biomass and total plant biomass in response to drought were stronger for woody plants than for herbs regardless of CO 2 treatments (p < 0.05; Figure 5a and b; Table S5). Compared with herbs, A of woody plants showed a larger response to eCO 2 (p = 0.04; Table S5).
However, woody plants and herbs showed no significant differences in their responses to the combination of drought and eCO 2 (p > 0.05; Figure 5c; Table S5). Under well-watered conditions, no significant difference in g s response to eCO 2 was detected between grass and forbs (p > 0.05; Figure S8a; Table S6); however, under drought conditions, a decrease in g s was observed for forbs but not for grass, although the difference was insignificant (p = 0.1; Figure S8b; Table S6). The drought × eCO 2 interaction was positive on g s for grass but was neutral for forbs (p = 0.005; Figure S8c; Table S6). There was a positive interaction between drought and eCO 2 on g s for shrubs rather than for trees, with the difference being significant (P = 0.05; Figure S9c; Table S7).

| The effect of experimental factors on plant responses
Experimental factors influenced the individual effects of drought and eCO 2 rather than their interactive effects on plants. Specifically, drought duration significantly affected the responses of Ψ predawn , WUE plant , A, above-ground biomass, below-ground biomass, total plant biomass and leaf area ( Figure 6; Table S8). The negative effect of drought duration on Ψ predawn at aCO 2 treatment weakened with increasing drought durations (p = 0.04; Figure 6a; Table S8). Similar patterns were observed for A at F I G U R E 4 Comparisons of the responses of C 3 plants and C 4 plants to elevated CO 2 (eCO 2 ) and its interaction with drought (D). (a) The effect of eCO 2 at well-watered treatment (W). (b) The effect of eCO 2 at D. (c) The interactive effect of eCO 2 and D. The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The asterisks indicate significant differences in the responses between C 3 plants and C 4 plants (*p < 0.05; **p < 0.01; ***p < 0.001). The numbers on the right represent the numbers of observations included. Refer to Figures 1-3 for the abbreviations F I G U R E 5 Comparisons of the responses of C 3 herbs and woody plants to drought (D) and its interaction with elevated CO 2 (eCO 2 ). (a) The effect of D at ambient CO 2 (aCO 2 ). (b) The effect of D at eCO 2 . (c) The interactive effect of eCO 2 and D. The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The asterisks indicate significant differences in the responses between herbs and woody plants (*p < 0.05; **p < 0.01; ***p < 0.001). The numbers on the right represent the numbers of observations included. Refer to Figures 1-3 for the abbreviations both aCO 2 (p = 0.03; Figure 6b; Table S8) and eCO 2 (p = 0.02; Figure 6b; Table S8). The drought-induced increase in WUE plant was observed in short term but disappeared in medium-or long-term drought under eCO 2 (p = 0.02; Figure 6b; Table S8). The mean effect size for aboveground biomass and below-ground biomass decreased the most in the long-term treatment regardless of CO 2 treatments (p < 0.05; Figure 6a,b; Table S8). The negative effect of drought at eCO 2 on total plant biomass was weakest in the short-term treatment (p = 0.04; Figure 6b; Table S8). The drought × eCO 2 interaction increased leaf area only in the short-term treatment (p = 0.03; Figure 6c; Table S8). Above-ground biomass, below-ground biomass and total plant biomass in experiments using GC and OTC responded more strongly to eCO 2 than those using FACE under well-watered conditions (p < 0.05; Figure 7a; Table S9). The decreases in Ψ predawn , A, above-ground biomass, below-ground biomass and total plant biomass in response to drought were stronger in pot than in field experiments regardless of CO 2 treatments (p < 0.05; Figure 8a,b; Table S10). The drought-induced reduction in g s was stronger in pot than in field experiments at eCO 2 (p = 0.04; Figure 8b; Table S10) rather than at aCO 2 (p > 0.05; Figure 8a; Table S10). Drought decreased Ψ midday to a greater extent in the experiments using big pots than in those using small pots at aCO 2 (p = 0.03; Figure S10a; Table S11), and there was positive interaction between drought and eCO 2 on Ψ midday in the experiments using big pots but not small pots (p < 0.001; Figure S10c; Table S11). The negative effects of drought on g s , A, above-ground biomass, below-ground biomass and total plant biomass were strongest in Type III drought manipulation at both aCO 2 and eCO 2 compared with Type I and Type II manipulations (p < 0.05; Figure S11a; Table S12).

| eCO 2 improves plant water relations without increasing SWC
We found that drought stress caused marked reductions in SWC regardless of CO 2 concentrations, resulting in a decrease in soil water availability. This is inconsistent with our expectation (H1) and some previous studies Robredo et al., 2007). It has been shown that the 44% decrease of g s in response to eCO 2 outweighed the 24% increase of leaf area, resulting in a slower soil water depletion under drought ; similar patterns of g s and leaf area responses to eCO 2 were observed in this study, which, however, did not translate into a higher SWC regardless of watering regimes, as recently reported by Jiang et al. (2021). A possible explanation is that eCO 2 enhances leaf temperature, which may partially counteract the reduction in plant transpiration due to the reduced g s caused by eCO 2 (Gray et al., 2016).
Additionally, it has been shown that the conservation of soil moisture induced by eCO 2 mainly occurred at shallow soil layers (Manderscheid et al., 2014). However, soil evaporation is mainly from the upper soil layer, which likely speeds up the consumption of the soil water saved by the reduced transpiration under eCO 2 (Manderscheid et al., 2018).
A process-based modelling also showed that eCO 2 did not increase soil moisture in spite of decreasing transpiration, but increased evaporation (Kellner et al., 2019). These findings suggest that changes in leaf area, leaf temperature and soil evaporation may collectively counteract the benefits of eCO 2 on soil water savings (Wilson et al., 1999).
Although no 'water saving effect' was detected, we found that eCO 2 improved leaf water status indicated by a higher LRWC under the combination of eCO 2 and drought stress. The improved leaf water status of droughted plants under eCO 2 has been shown to be accompanied by increases in Ψ midday and Ψ predawn (Robredo et al., 2007). However, a less negative Ψ midday rather than Ψ predawn was observed in the present study. This is consistent with the finding of Atwell et al. (2007) that the improved water status only occurred during daylight hours, suggesting that the effect of eCO 2 was a result of decreased transpiration rather than increased SWC (Field et al., 1995). Given that a higher eCO 2induced increase in LRWC was observed only when drought occurred, the decreased sensitivity of g s to eCO 2 caused by drought suggests that there may be other mechanisms that contributed to the improved leaf water status. For example, root biomass production was preferentially stimulated by eCO 2 when plants were subjected to water stress, leading to a higher R/S. This proportionally larger investment of C in root growth may allow plants to have more extensive root systems and improve F I G U R E 6 Effects of drought durations on plant responses to drought (D) and its interaction with elevated CO 2 (eCO 2 ). (a) The effect of D at ambient CO 2 (aCO 2 ). (b) The effect of D at eCO 2 . (c) The interactive effect of eCO 2 and D. The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The asterisks indicate significant differences in the responses among the drought durations (*p < 0.05; **p < 0.01; ***p < 0.001). The numbers on the right represent the numbers of observations included. Refer to Figures 1-3 for the abbreviations their water acquisition (Idso & Idso, 1994;Wullschleger et al., 2002).
These findings suggest that stomatal control and morphological adjustments jointly improve leaf water status under eCO 2 and drought (Jiang et al., 2021), and eCO 2 consequently mitigates the impacts of drought

| eCO 2 alleviates adverse effects of drought on photosynthesis
We found that the drought-induced reduction in g s was concurrent with the increase in leaf ABA concentration, suggesting that ABA may be involved in regulating stomatal closure under drought stress (Comstock, 2002). However, the drought-induced enhancement in leaf ABA concentration was less pronounced when plants were exposed to eCO 2 . This negative interaction between drought and eCO 2 has been suggested to delay stomatal response to drought , which was supported by our finding that the response of g s to drought was reduced by eCO 2 . Similar responses have been observed in Lycopersicon esculentum (Liu et al., 2019), Fagus sylvatica and Castanea sativa (Heath, 1998). The smaller reduction in g s caused by drought at eCO 2 was also in accordance with the aforementioned finding that LRWC and Ψ midday were increased by eCO 2 at drought conditions. This result indicates that eCO 2 indirectly impacted leaf F I G U R E 7 Effects of experimental protocols on plant responses to elevated CO 2 (eCO 2 ) and its interaction with drought (D). (a) The effect of eCO 2 at well-watered treatment (W). (b) The effect of eCO 2 at D. (c) The interactive effect of eCO 2 and D. The effect size is calculated as a percentage response (%). FACE, free-air CO 2 enrichment; GC, growth chamber; OTC, open top chamber. The error bars represent 95% confidence intervals. The asterisks indicate significant differences in the responses among the experimental protocols (*p < 0.05; **p < 0.01; ***p < 0.001). The numbers on the right represent the numbers of observations included. Refer to Figures 1-3 for the abbreviations F I G U R E 8 Effects of growth conditions on plant responses to drought (D) and its interaction with elevated CO 2 (eCO 2 ). (a) The effect of D at ambient CO 2 (aCO 2 ). (b) The effect of D at eCO 2 . (c) The interactive effect of eCO 2 and D. The effect size is calculated as a percentage response (%). The error bars represent 95% confidence intervals. The asterisks indicate significant differences in the responses between the growth conditions (*p < 0.05; **p < 0.01; ***p < 0.001). The numbers on the right represent the numbers of observations included. Refer to Figures 1-3 for the abbreviations water status through its effect on g s , which, in turn, affected the response of g s to drought. However, eCO 2 changed the magnitude of g s response to drought but not the direction, suggesting that SWC may play an important role in stomatal control, and that stomata respond to the factors that influence plant water status (Buckley, 2019).
Following the decreased g s under drought, there was a decrease in C i , suggesting that drought imposed stomatal limitation on A (Flexas & Medrano, 2002). In contrast, eCO 2 alleviated stomatal limitation by stimulating C i , resulting in a more pronounced enhancement in A under drought treatment than under well-watered treatment; this provides evidence for the 'low C i effects' hypothesis Kelly et al., 2016). In addition, the less sensitive of g s to drought for plants growing under eCO 2 may be another reason for the positive interaction between drought and eCO 2 on A. However, the eCO 2 -induced stimulation of biomass was independent of water availability. This is likely because the growth response to eCO 2 increases with decreasing soil moisture only when eCO 2 produces relatively wet soil (Fatichi et al., 2016;Morgan et al., 2004;Ottman et al., 2001). Given that no 'water saving effect' was detected in this study, it is logical that there was no positive interaction between drought and eCO 2 on biomass.

| eCO 2 increases WUE, but the magnitude is scale dependent
At the leaf level, iWUE increased under eCO 2 as a consequence of the reduced g s and enhanced A, in line with previous meta-analyses (Ainsworth & Long, 2005;Wang & Wang, 2021). It has been suggested that the impact of eCO 2 on WUE was lower in plants under drought (De Kauwe et al., 2013;Robredo et al., 2007). However, drought did not affect the response of iWUE to eCO 2 , which, combined with the finding of the proportional increase in iWUE with eCO 2 regardless of water availability, supports the optimal stomatal behaviour theory that g s and A are well coupled to maximize C uptake and minimize water loss (Medlyn et al., 2011;Walker et al., 2021;Wang & Wang, 2021). Similarly, it has been observed that when the eCO 2 -induced stimulation of A was greatest, the reduction in g s was smallest, suggesting a tight coupling between A and g s (Pastore et al., 2019). Such an iWUE response to eCO 2 has been suggested to be regulated by three mechanisms, that is, maintaining a constant C i , C i − C a and C i :C a (Saurer et al., 2004). The present meta-analysis extends previous empirical findings and model simulations by showing that A and g s are regulated in a way to keep the C i :C a constant in response to eCO 2 , which is independent of soil water availability; this suggests a consistent and moderate contribution of eCO 2 to the increase in iWUE (Ainsworth & Long, 2005;De Kauwe et al., 2013;Peñuelas et al., 2011).
At the individual level, there was also an eCO 2 -stimulated WUE plant , suggesting that the higher biomass production was not accompanied by a proportional increase in water use; this is in line with the large-scale evidence that increased terrestrial C uptake by eCO 2 does not cause an enhancement in water use because of the increased WUE (Cheng et al., 2017). It has been shown that eCO 2 improved WUE plant to a greater extent in drought treatment than in well-watered treatment (Qiao et al., 2010). However, no interaction between eCO 2 and drought was detected in this study, likely because of the lack of positive interaction between eCO 2 and drought on plant biomass.
There is evidence that the sensitivity of WUE to eCO 2 decreased from leaf to plant levels (Centritto et al., 1999;Kelly et al., 2016;Knauer et al., 2017), which was confirmed by the present study. This is likely because additional feedbacks may play a role in scaling up iWUE to the whole-plant level (Centritto et al., 1999;De Kauwe et al., 2013;Field et al., 1995;Knauer et al., 2017). First, the aforementioned larger leaf area under eCO 2 would reduce the physiological effects of eCO 2 on plant water use (Field et al., 1995).
Second, the response magnitude of plant/stand-scale WUE to eCO 2 also depends on the coupling between the leaf and the atmosphere (De Kauwe et al., 2013). For example, Kelly et al. (2016) attributed the discrepancy between the responses of iWUE and WUE plant of Eucalyptus seedlings to eCO 2 to the weak coupling between plants and the surrounding air. Third, the difference in leaf-to-air vapour pressure deficit and responses of A and g s to eCO 2 can change vertically within canopies owing to changes in light availability (Barton et al., 2012).

| Factors affecting plant responses to drought and eCO 2
The interactive effect of drought and eCO 2 on A may depend on photosynthetic pathway (Leakey et al., 2006). Our study showed that the photosynthetic advantage of C 3 plants over C 4 plants under eCO 2 diminished with the onset of drought stress, which is consistent with the finding that the response of A to eCO 2 was more positive under reduced rainfall compared with ambient rainfall for C 4 grass but not for C 3 grass (Pastore et al., 2020). The distinct responses of C 3 and C 4 plants to the combination of eCO 2 and drought may be associated with the unique CO 2 -concentrating mechanism of C 4 plants. Compared with C 3 plants, the initial slope of the A/C i curve of C 4 plants is much steeper, and A is CO 2 saturated at a lower C i and thus is less responsive to eCO 2 (Leakey, 2009). However, when C 4 plants are exposed to drought, the reduction in g s may decrease the operating C i to a value below the inflexion point of the A/C i curve, and A becomes more sensitive to eCO 2 . Similarly, drought provoked a more pronounced WUE plant response of C 4 plants to eCO 2 . This finding supports a model simulation that a stronger enhancement in WUE plant of Zea mays caused by eCO 2 occurred in drought treatment than in the wet treatment (Kellner et al., 2019). The benefits of eCO 2 on A and WUE plant of C 4 plants under water-limited environments may explain why the biomass of C 4 plants was enhanced only when eCO 2 and drought were combined (Leakey et al., 2006;Manderscheid et al., 2014;Ottman et al., 2001).
There were also significant differences in plant response to eCO 2 and drought among C 3 functional groups. The negative effects of drought on plant biomass of herbs were less than those of woody plants, which is generally consistent with the viewpoint that compared with woody plants, herbs have more strategies (e.g. dehydration escape; Kooyers, 2015) to cope with drought and thus are more drought tolerant (Volaire, 2018). Contrarily, woody plants exhibited significantly greater increases in A in response to eCO 2 than herbs, which is in line with previous findings that woody plants responded more strongly to eCO 2 (Ainsworth & Long, 2005).
Unexpectedly, the interaction between drought and eCO 2 on plants hardly changed with experimental factors such as drought manipulation type, experimental protocol, growth condition, drought duration, etc., which, however, mediated the individual effects of drought and eCO 2 on plants. For example, we found that the physiology and biomass production were more strongly constrained by drought for plants growing in pots than those growing in field. This is consistent with a previous meta-analysis that the effects of precipitation changes on root biomass in pot experiments did not mirror those in field experiments . Our finding implies that plants growing in pots were less drought tolerant than those growing in field, likely because pot size impacted root growth and development (Poorter et al., 2012); rooting volume, fine-root area and activity determine the capacity of the root system to take up water (Wullschleger et al., 2002). In addition, the negative effect of drought on plant biomass worsened as drought prolonged, possibly because the plant growth became C limited over long-term drought (Duan et al., 2013). Additionally, experimental protocol affected plant responses to eCO 2 . Specifically, plant biomass in FACE experiments was less responsive to eCO 2 , similar to the result by de Graaff et al. (2006). We also found that drought caused the greatest reductions in g s , A and biomass for Type III manipulation; this suggested that plant physiology and growth were severely impaired when the water supply was totally withheld (Type III), whereas recurrent mild droughts (Type II) may increase plant drought resistance (Backhaus et al., 2014;Bréda et al., 2006).

| CON CLUS IONS
To detect whether eCO 2 can alleviate the negative effects of drought stress, we performed a worldwide synthesis on the interactive effects of eCO 2 and drought on plant water status, photosynthesis, WUE, biomass production and allocation. Our analysis showed that eCO 2 little affected SWC, but improved leaf water status under drought conditions (e.g. a higher LRWC and a less negative Ψ midday ) by reducing g s and increasing R/S. Elevated CO 2 enhanced WUE regardless of soil water availability, which was jointly driven by a lower g s response (associated with leaf ABA levels) and a higher A response to eCO 2 under drought, consistent with the optimal stomatal behaviour, but the magnitude of the eCO 2 -induced enhancement in WUE decreased from the leaf to individual scales. eCO 2 reduced the magnitude of the effect of drought on g s , but did not change the direction. The lower g s under drought caused stomatal limitations on A, while eCO 2 alleviated stomatal limitations by increasing C i , which resulted in a greater A response to eCO 2 and drought, supporting the 'low C i effect' hypothesis. However, the magnitude of the increase in plant biomass caused by eCO 2 did not vary with water availability. The advantages of eCO 2 on C 3 plants over C 4 plants under well-watered conditions diminished under drought conditions. Compared with C 3 herbs, drought caused a greater reduction in biomass of woody plants. The negative effect of drought on plant biomass increased as drought prolonged. Plants growing in pots were less drought tolerant than those growing in field. The eCO 2 -induced increase in biomass was observed in growth chamber and OTC experiments rather than in FACE experiments. These findings suggest that eCO 2 can alleviate the adverse impacts of drought by improving plant water status and A; they enhance our understanding of plant responses to and feedbacks on global changes.
Nevertheless, we realized several limitations in this study.
First, although we found that eCO 2 improved plant water status, of which the exact mechanisms remain uncertain. One potential candidate is that eCO 2 may affect the rooting depth and vertical distribution of roots, which determine the water acquisition capacity of roots (Nadal-Sala et al., 2021;Wullschleger et al., 2002).
Second, eCO 2 often increases non-structural carbohydrates, which may allow plants to increase osmotic adjustment and maintain a higher water potential (Miranda-Apodaca et al., 2018). Third, although normalizing plant responses to eCO 2 with the magnitude of CO 2 treatments did not change the directions of the responses, it increased the magnitudes by 38.8%-42.0% on average; nevertheless, the current dataset denied the normalization to drought and its interaction with eCO 2 . Fourth, eCO 2 can also impact plant water supply through its effect on plant hydraulic conductance, with the specific effects depending on species and plant growth form (Domec et al., 2017). However, few data are available for synthesizing interactive effects of eCO 2 and drought on those aspects; the number of observations for SWC was also smaller than those of plant physiological variables, which may limit the power of this meta-analysis. Clearly, these limitations call for more studies, particularly on exploring responses of plant root characteristics (e.g. root length, root distribution), hydraulics (e.g. osmotic adjustment, plant hydraulic architecture) and SWC to the combination of eCO 2 and drought.

AUTH O R CO NTR I B UTI O N S
Zhaoguo Wang and Chuankuan Wang designed the study.
Zhaoguo Wang collected and analysed the data. Zhaoguo Wang and Chuankuan Wang drafted the manuscript. Zhaoguo Wang, Chuankuan Wang and Shirong Liu were involved in the revision.

ACK N OWLED G EM ENTS
We thank all the researchers whose data were used in this metaanalysis, Dr. Alessio Collalti and the anonymous reviewers for their valuable comments. This work was financially supported by the National Key Research and Development Program of China (2021YFD2200401).

CO N FLI C T O F I NTE R E S T
The authors declare there are no conflicts of interest.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/1365-2745.13988.

DATA AVA I L A B I L I T Y S TAT E M E N T
The database used in this meta-analysis is available from the Dryad Digital Repository https://doi.org/10.5061/dryad.zs7h4 4jcz (Wang et al. 2022).