Physiological responses to low CO2 over prolonged drought as primers for forest–grassland transitions

Savannahs dominated by grasses with scattered C3 trees expanded between 24 and 9 million years ago in low latitudes at the expense of forests. Fire, herbivory, drought and the susceptibility of trees to declining atmospheric CO2 concentrations ([CO2]a) are proposed as key drivers of this transition. The role of disturbance is well studied, but physiological arguments are mostly derived from models and palaeorecords, without direct experimental evidence. In replicated comparative experimental trials, we examined the physiological effects of [CO2]a and prolonged drought in a broadleaf forest tree, a savannah tree and a savannah C4 grass. We show that the forest tree was more disadvantaged than either the savannah tree or the C4 grass by the low [CO2]a and increasing aridity. Our experiments provide insights into the role of the intrinsic physiological susceptibility of trees in priming the disturbance-driven transition from forest to savannah in the conditions of the early Miocene. The ancient expansion of savannahs has long been examined through models and palaeorecords, but this new experiment combining CO2 and drought finds the physiological mechanisms priming the forest-to-savannah transition.

, but direct comparative experimental evidence for tree and grass responses to variations in [CO 2 ] a and drought is lacking.
Here we investigate the physiological responses of present-day savannah and forest species to concurrent variations in [CO 2 ] a and water availability. We hypothesized, first, that decreasing [CO 2 ] a would reduce CO 2 assimilation rates more in C 3 trees than in C 4 grasses during drought and would increase water loss from leaves, amplifying drought effects (H 1 ); second, that because savannah trees evolved under declining [CO 2 ] a and increasing aridity, they would be less disadvantaged than forest trees by sub-ambient [CO 2 ] a under prolonged drought (H 2 ); and, third, that because starch is the main reserve for resprouting, the capacity of trees and grasses to recover from drought would depend on the content of starch in leaves and roots (H 3 ).
To evaluate our three physiology-based savannah evolution hypotheses, we undertook replicated comparative experimental trials across three species of representative contrasting functional types: Celtis africana (N. L. Burm.), a deciduous broadleaf forest tree common in southern Africa that represents plants of the fire-sensitive forests and woodlands that dominated before C 4 grassy savannah expansion 9 ; Vachellia karroo (Hayne) (Acacia), a nitrogen-fixing tree, typical of open savannahs; and Eragrostis curvula ((Schrad.) Nees), a drought-tolerant C 4 grass common to frequently burned or grazed mesic savannahs 4,24 . The plants were grown in replicated controlled environments under 200, 400 or 800 ppm [CO 2 ] a , and a controlled drought was imposed over five months by reducing watering in four steps from 80% to 30% of gravimetrically determined pot capacity, followed by a month-long re-watering back to 80%. We characterized canopy transpiration, leaf gas exchange, photosynthetic physiology and starch content at all [CO 2 ] a -watering level combinations, providing a detailed description of the physiological responses of savannah species to the range of [CO 2 ] a expected for past, current and future scenarios. of growth [CO 2 ] a at all watering levels ( Fig. 1; see Extended Data Table 1 for the P values and details of the statistical analysis). In the grass, E plant was higher at 200 ppm ('sub-ambient') than at 800 ppm ('elevated') [CO 2 ] a , under both well-watered and drought conditions (Fig. 1). All species responded to drought with an immediate decline in E plant , which was steeper in the savannah species V. karroo and E. curvula than in C. africana. Under well-watered conditions, leaf relative water content (RWC) consistently scaled with growth [CO 2 ] a , but the differences were small and were significant in E. curvula only. Drought resulted in an immediate decline in RWC in the savannah species; the reduction was delayed until 40% and 30% watering levels in C. africana, but RWC then dropped at markedly lower levels, especially at sub-ambient [CO 2 ] a .
For all species under well-watered conditions, midday water potential (Ψ leaf ) was significantly higher at elevated and ambient than at sub-ambient [CO 2 ] a ( Fig. 2). At the onset of drought, significant reductions in Ψ leaf and predawn leaf water potential (Ψ pd , a proxy for soil water potential) occurred for savannah species at all [CO 2 ] a , while the response was delayed until 50% watering in C. africana.
For most combinations of species and [CO 2 ] a , re-watering recovered E plant , RWC, Ψ leaf and Ψ pd to values similar to or higher than those for well-watered plants, except  Photosynthetic and stomatal responses. In the trees under well-watered conditions, operational photosynthetic rate (A op ) was higher at elevated than at sub-ambient [CO 2 ] a , but A op was insensitive to [CO 2 ] a levels in well-watered E. curvula (Fig. 3). The response of operational stomatal conductance (g s ) to [CO 2 ] a under well-watered conditions was similar in the savannah species V. karroo and E. curvula, progressively decreasing with growth [CO 2 ] a . In the savannah species, A op and g s responded faster to drought than in C. africana, with large decreases in A op and g s as soon as watering was reduced to 60%, especially at ambient and sub-ambient [CO 2 ] a . After the initial drop, g s stabilized in E. curvula but continued to decrease in V. karroo. In contrast, C. africana had a gradual reduction in A op that became significant only at 50%, 40% and 30% watering for high, ambient and sub-ambient [CO 2 ] a , respectively. In C. africana, the response of g s to drought was significant only at elevated [CO 2 ] a . Re-watering led to the recovery of A op and g s to initial levels in all species and [CO 2 ] a , except for A op in C. africana and g s in V. karroo, which remained low at sub-ambient [CO 2 ] a .
Photosynthetic capacity. In E. curvula, all the photosynthetic parameters (R light , Y(CO 2 ) LL , GA sat and LCP, defined below; Fig. 4) derived from curves measuring the response of leaf-level CO 2 assimilation (A) to increasing photosynthetic photon flux density (PPFD) (A-PPFD curves) (Extended Data Fig. 3) were independent of growth [CO 2 ] a . The trees grown at elevated [CO 2 ] a had higher respiration in the light (R light ) and light-saturated gross assimilation rate (GA sat ) than at sub-ambient [CO 2 ] a (Fig. 4). V. karroo grown at elevated [CO 2 ] a had higher initial quantum yield for CO 2 fixation (Y(CO 2 ) LL ) for all watering levels.
All species responded to drought with a concerted downregulation of photosynthetic parameters. However, only C. africana reduced R light and was able to maintain constant Y(CO 2 ) LL despite drought, resulting in a progressive reduction in the light compensation point (LCP) as drought progressed. In contrast, drought decreased Y(CO 2 ) LL in E. curvula, which negatively impinged on the LCP. In all species and at all [CO 2 ] a levels, GA sat decreased with drought. The steepest decrease was for V. karroo, which, at 40% watering, reduced GA sat to an average across [CO 2 ] a levels of 12% of the initial values, compared with 44% and 53% for E. curvula and C. africana, respectively.
In C. africana, all the photosynthetic parameters (CE, V cmax , Γ and A sat , defined below; Fig. 5 and Extended Data Fig. 4) derived from curves measuring the response of A to [CO 2 ] in the sub-stomatal cavity (C i ) (A-C i curves), measured under a PPFD of 1,500 μmol m −2 s −1 , were independent of growth [CO 2 ] a . V. karroo had higher carboxylation efficiency (CE) and CO 2 -saturated assimilation rate (A sat ) when grown at elevated [CO 2 ] a (Fig. 5)  Responses of E plant (top) and RWC (bottom) to watering levels and for three levels of growth [CO 2 ] a in the forest broadleaf tree C. africana (left), the savannah tree V. karroo (centre) and the C 4 savannah grass E. curvula (right). The plants were grown in controlled-environment chambers at 200 ppm (Sub), 400 ppm (Amb) or 800 ppm (ele) [CO 2 ] a , and they were subjected to progressive month-long decreasing watering levels of 80%, 60%, 50%, 40% and 30% of pot capacity followed by re-watering (Re) back to 80% (indicated by different shades of grey). The values are means ± s.e. (n = 4). Horizontal differences (across watering levels within a given [CO 2 ] a , not across CO 2 levels; P < 0.05) are indicated by different letters. Within waterings, differences between sub-ambient and elevated are indicated with §, between sub-ambient and ambient with †, and between elevated and ambient with ‡ (P < 0.05). When symbols appear in the upper left corner, they apply to all waterings. The statistical details and P values are given in extended Data Table 1      ] a , and they were subjected to progressive month-long decreasing watering levels of 80%, 60%, 50%, 40% and 30% of pot capacity followed by re-watering back to 80% (shades of grey). The values are means ± s.e. (n = 4). The different letters indicate horizontal significant differences (P < 0.05) across watering levels within a given [CO 2 ] a , not across CO 2 levels. The symbols indicate differences (P < 0.05) between sub-ambient and elevated ( §), sub-ambient and ambient ( †) or elevated and ambient ( ‡); when shown in the upper left corner, they apply to all waterings. The statistical details and P values are given in extended Data Table 1. differences were delayed until 50% watering in the other two species. Γ increased with drought only in V. karroo. The response of the photochemical integrity of photosystem II as indicated by F v /F m (the ratio between variable fluorescence and maximum fluorescence) is shown in Extended Data Fig. 6. E. curvula maintained F v /F m throughout the experiment, while F v /F m followed the general trend of A op in the trees: dropping at drought onset in V. karroo and later in C. africana.
Re-watering resulted in the recovery of all photosynthetic parameters to initial levels in V. karroo, and all parameters except Y(CO 2 ) LL and LCP in E. curvula. However, in C. africana, R light , GA sat , LCP and V cmax did not recover (Figs. 4 and 5).
Leaf and root starch content. In E. curvula, the starch content in both leaves and roots was independent of growth [CO 2 ] a (Fig. 6). In V. karroo, root starch content was not affected by growth [CO 2 ] a , were measured at the growth [CO 2 ] a on randomly selected plants during each month-long water-level interval and were used to derive the following parameters: R light (the y-intercept of a hyperbola fitted to A-PPFD curves), Y(CO 2 ) LL (the initial slope), GA sat (the asymptote) and LCP (the y-intercept; that is, PPFD when GA = R light ). The values are means ± s.e. (n = 4). The letters indicate horizontal significant differences (P < 0.05) across watering levels within a given [CO 2 ] a of 200 (black upright font), 400 (italic font) or 800 ppm (light grey font). If the differences apply to all [CO 2 ] a , only one set of letters are displayed. The symbols indicate vertical differences (P < 0.05) between sub-ambient and elevated ( §), sub-ambient and ambient ( †) or elevated and ambient ( ‡); when shown at the left, they apply to all waterings. The statistical details and P values are given in extended Data In the trees, leaf starch content decreased with drought, rapidly in V. karroo (significant decreases occurred at 60-50% watering) and gradually in C. africana (significant effects were delayed until 30% watering). Conversely, in the C 4 grass E. curvula, leaf starch content did not decrease with drought. In the savannah tree and grass, leaf starch content increased at lower watering levels at sub-ambient and ambient [CO 2 ] a .
Root starch content slowly decreased with drought in V. karroo, but it was independent of watering in the other species. After re-watering, leaf starch content recovered to initial values in the trees, but it became significantly lower than at any other period in the grass. Root starch remained low in V. karroo.

Effect of [CO 2 ] a over prolonged drought.
To compare the effect of growth [CO 2 ] a over a season-long drought across species, we computed the difference between the natural logarithms of values at either sub-ambient or elevated [CO 2 ] a and those at ambient [CO 2 ] a for each variable. The change in the natural logarithms of variables is independent of the initial value and unit of the variable, and therefore allows for direct comparison across species (see the note to Table 1 for more details). We tested whether the response differed between trees and the C 4 grass (H 1 ) or between the two trees (H 2 ).  Fig. 4) were measured on randomly selected plants during each month-long water-level interval and were used to derive the following parameters: Ce (the initial slope of a fitted hyperbola); V cmax for the C 3 trees and V pmax for the C 4 grass, which were mechanistically derived; Γ (the y-intercept; that is, C i when A = 0); and A sat (the asymptote, reflecting the potential for electron transport). The values are means ± s.e. (n = 4). The letters indicate significant differences (P < 0.05) across watering levels and apply to all [CO 2 ] a . The symbols indicate vertical differences (P < 0.05) between sub-ambient and elevated ( §), sub-ambient and ambient ( †) or elevated and ambient ( ‡); when shown at the left, they apply to all waterings. The statistical details and P values are given in extended Data Table 1. No values indicates that there were no viable (trees) or measurable (grass) leaves. In trees leaves were dropped in grasses leaves were rolled.
Growth at sub-ambient [CO 2 ] a decreased g s , A op , GA sat and leaf starch more in the trees than in the grass and decreased Γ more in the grass than in the trees (Table 1). Comparing the savannah species, sub-ambient [CO 2 ] a decreased all photosynthetic indicators more in the savannah tree, V. karroo, than in the savannah grass, E. curvula. The trees responded to sub-ambient [CO 2 ] a by decreasing A op and LCP, but these decreases were larger in C. africana than in V. karroo. Interestingly, this was accompanied by a larger decrease in g s in V. karroo than in C. africana, indicating higher non-stomatal limitation in C. africana. The response accompanied by an increase in Ci in C. africana [Extended Data Fig. 5] to elevated [CO 2 ] a was similar between the trees and the grass, but A op and GA sat increased more in V. karroo than in C. africana and E. curvula.

Discussion
Our experimental design combined three [CO 2 ] a levels spanning the likely range of values expected for past and future scenarios (200, 400 and 800 ppm) with a gradual stepped reduction in water supply (80% to 30% of pot capacity) followed by re-watering. This allowed for the investigation of the physiological responses of acclimation with unprecedented temporal resolution, avoiding confounding effects of rapid senescence that occur when plants are deliberately killed in irrigation suspension experiments 25 .
Our first hypothesis (H 1 ) stated that during a prolonged drought, sub-ambient [CO 2 ] a would reduce assimilation more in trees than in grasses, and this was supported by the experimental evidence: sub-ambient [CO 2 ] a reduced A op and GA sat more in trees than in the C 4 grass E. curvula; however, contrary to what we thought, over the course of drought, sub-ambient [CO 2 ] a increased g s more in E. curvula than in the trees ( Table 1). The response to drought was found to be time-dependent. In C 4 grasses, short-term dips in leaf water potential between days 26 or around midday 27 cause rapid declines in assimilation that are mainly due to non-stomatal limitation 28 . Here we found that, after an initial reduction, E. curvula stabilized A op and g s , while photosynthesis continued to decline in the trees. We reason that E. curvula may take days to acclimate to drought; when water potential falls fast, non-stomatal limitations would reduce assimilation to the point where avoidance becomes advantageous. Indeed, from 60% watering onwards, E. curvula rolled its leaves in the evening of the day after watering, and increasingly earlier at lower watering levels, then promptly unfurled after watering. In contrast, the trees started dropping their leaves at 50% watering, and by 30% watering they were almost completely defoliated.
Both leaf shedding and leaf rolling effectively reduce canopy evapotranspiration, conferring protection against hydraulic damage 29 . While leaf shedding is ubiquitous, rolling is more common in spontaneous and cultivated grasses and cereals [30][31][32] . Rolled leaves remain in situ, meaning that essential resources (such as nitrogen and phosphorous) are retained in live tissue rather than relocated or lost through leaf senescence. Unfurling operates over hourly timescales, allowing rapid exploitation of post-or mid-drought soil water pulses. Conversely, recovery after leaf shedding requires growing new leaves, imposing a lag time of weeks or months. Redmann 33 proposed that rolling amphistomatous leaves would be selected for in habitats where diurnal water supply and demand fluctuate widely. These mechanisms explained a rapid 90% increase in assimilation in a C 4 savannah grass following a mid-drought rainfall event compared with only a 22-26% increase in oak trees 34   ] a , and they were subjected to progressive month-long decreasing watering levels of 80%, 60%, 50%, 40% and 30% of pot capacity followed by re-watering back to 80% (shades of grey). The values are means ± s.e. (n = 4). The letters indicate significant differences (P < 0.05) across watering levels within a given [CO 2 ] a , not across CO 2 levels. The symbols indicate differences (P < 0.05) between sub-ambient and elevated ( §), sub-ambient and ambient ( †) or elevated and ambient ( ‡); when shown in the upper left corner, they apply to all waterings. The statistical details and P values are given in extended Data Table 1. No values indicates that there were no viable (trees) or measurable (grass) leaves. In trees leaves were dropped in grasses leaves were rolled. grasses a growth advantage over trees relying on regrowth 29 . This advantage would be more pronounced under sub-ambient [CO 2 ] a , as the photorespiratory and evaporative losses of C 3 trees accelerate senescence 35,36 .
Our second hypothesis (H 2 ), that the savannah tree V. karroo would be less disadvantaged at sub-ambient [CO 2 ] a than the forest tree C. africana, was also supported by our experimental evidence (Table 1). However, the timing of the response to drought differed between the trees. V. karroo rapidly closed its stomata (Fig. 3) (consistent with a previously described isohydric behaviour 37 ), reduced transpiration (Fig. 1), dropped leaves and accumulated starch (Fig. 6). Conversely, C. africana maintained both E plant and leaves for longer ( Figs. 1 and 3), but at the expense of marked leaf dehydration as drought intensified at sub-ambient [CO 2 ] a (Fig. 1). Upon re-watering, V. karroo fully recovered A op and photosynthetic potential, whereas C. africana did not recover leaf RWC at elevated [CO 2 ] a , A op under sub-ambient [CO 2 ] a ( Figs. 1 and 3) and four of the eight photosynthetic parameters (GA sat , V cmax , R light and LCP) (Figs. 4 and 5), further supporting H 2 .
By delaying leaf senescence, C. africana may have traded its capacity to remobilize nutrients from leaves to storage organs, and this, in turn, may have affected its capacity to recover, particularly under sub-ambient [CO 2 ] a . C. africana maintained dehydrated leaves on the tree for some time before shedding. Contrasting strategies of rapid versus delayed leaf shedding were also observed in closely related savannah and forest eucalyptus species 38 , potentially underpinned by the advantage of shading out competitors 21,39 . Furthermore, C. africana reduced leaf respiration, which resulted in maintaining quantum yield and in a substantial reduction of the LCP (Fig. 4). This is critical under low light 40 because it helps saplings maintain a net positive carbon balance in the shady forest understory 41 . Overall, this adaptive shade strategy may be successful during mild, short droughts 6 , but not for longer periods.
Our third hypotheses (H 3 ), that the capacity to recover from drought depends on the content of starch in roots and leaves, was not supported by our results; recovery was independent of the starch content of both organs. However, we found that leaf starch content might be related to drought resilience. In both E. curvula and V. karroo under sub-ambient [CO 2 ] a , leaf starch content at 30% and 40% watering was significantly higher than at higher watering levels, but such an increase was not observed in C. africana (Fig. 6). This change in leaf starch content was accompanied by more negative Ψ in the savannah tree and grass than in the forest tree C. africana (Fig. 2). Starch hydrolysis may play a central role in cavitation repair and osmotic adjustment following drought injury [42][43][44] . Concentrating starch in leaves could thus have increased the operational safety margins against hydraulic failure for the savannah species under drought. Our observations are consistent with reports showing the accumulation of non-structural carbohydrates in V. karroo leaves under drought conditions 37 . In montane grasslands under ambient [CO 2 ] a , the proportion of recent assimilate allocated to root storage increased during drought 45 . In this view, starch probably functioned as insurance against drought at the expense of growth, rather than a sink for carbon overspill 3,46-51 . Finally, our data do not support the suggestion that leaves under high [CO 2 ] a accumulate starch to levels that inhibit photosynthetic performance 18 . Starch content increased with [CO 2 ] a only within C. africana roots, but this was associated with an increase in A op .
Physiological responses contribute to ecosystem shifts from forests to savannahs and vice versa. Our results show that forest trees were disadvantaged by low [CO 2 ] a and increasing aridity, conditions that occurred during the early Miocene. Uncertainties in the lower bound of Miocene [CO 2 ] a would probably affect the magnitude of plant responses, and across a lower gradient in [CO 2 ] a , the magnitude of treatment effects could be diminished 52 . Trees carry a carbon tax on large woody structures, both structural and for maintenance, which is avoided by herbaceous grasses. Low [CO 2 ] a and drought could result in the death of branches and trees through carbon starvation or hydraulic failure, but generally these occur under exceptional levels of stress 53 . It has been shown that drought and low [CO 2 ] a were not sufficient by themselves to tip the transitional balance to open ecosystems 54 , but they would probably increase the intrinsic susceptibility of trees to the effects of disturbance. This may have played a role in the expansion of grassy savannahs at the expense of forests under low [CO 2 ] a 9 some 20 million years ago 1 . Under low [CO 2 ] a , trees grow slower and allocate less biomass to both physical (thorns) 55 and chemical defences 36 , accelerating canopy opening by browsers 56 . While forest grasses seldom produce enough biomass to burn, opening environments would accumulate more grass biomass, fuelling deeper penetration of fires. The competition for water and nutrients would intensify, and slow-growing trees would be less likely to grow tall enough to escape destruction by fire (escaping the 'fire trap'). This, in turn, would reduce sapling recruitment to mature trees 57 , conditioning the ecosystems to entrain positive feedbacks involving fires 2 . It has been suggested that during early stages of forest-grassland transition, local factors including soil properties, rainfall intensity and topography were critical in shaping the competition between grasses and tree saplings 5,58-62 . In these opening habitats, quick-responding trees such as V. karroo would progressively displace forest trees such as C. africana. The shift in the proportion of C 3 and C 4 grasses could have been driven by contrasting responses of C 3 and C 4 grasses to drought 63 , ultimately giving these emerging ecosystems the shape of modern savannahs. Differences in the timing of the response 26 suggest that C 4 grasses evolved different mechanisms of avoidance and tolerance of drought than C 3 grasses (in addition to those described above, these mechanisms include faster-responding stomata 64 , enhanced water delivery to leaves 26,65 and low stomatal conductance 66 ).
Similar interactions between physiology and disturbance may operate at increasing [CO 2 ] a . Ecosystem models predict that elevated [CO 2 ] a will shift areas that could currently support multiple stable biomes into tree-dominated areas 67 . Higher C 3 assimilation under elevated [CO 2 ] a allows tree saplings to grow taller, increasing their probability of escaping the fire trap 68 . Higher carbon assimilation allows greater allocation to mechanical 55 and chemical 36 defences and effective cavitation repair. Furthermore, reduced transpiration translates into water savings at the ecosystem level 69 and may reduce flammability. These factors all drive the thickening of savannahs 39,48,70 , which is predicted to continue under future climatic scenarios 67 . In the understory of progressively closing ecosystems, shade-tolerant saplings such as C. africana will have an advantage and could be preferentially recruited. Growing forest trees would eventually outshade savannah trees 71 , transitioning back to closed, less flammable tropical forest 72 .

Methods
Plants and growth conditions. Seedlings of V. karroo were obtained from the Desert Legume Program (Tucson, AZ, USA), accession number 900474, collected outside the Mountain Zebra National Park, northwest of Cradock (Eastern Cape, South Africa); those of C. africana were obtained from Silverhill Seeds (Cape Town, South Africa); and a clone of E. curvula was obtained from the Germplasm Resources Information Network (United States Department of Agriculture, Washington, DC, USA), accession number PI-155434. The plants were grown in 2.5 dm 3 pots filled with a loam soil that has a gradual decline in soil water potential when dried (Extended Data Fig. 2). The plants were randomly distributed among six controlled-environment growth chambers at duplicated [CO 2 ] a levels of 200, 400 or 800 ppm, rotated weekly within and monthly between cabinets. The temperature was set at 26:17 °C, and the relative humidity was 70:50% (day:night).
The trees and grasses were initially grown for 18 and 6 months, respectively, while being watered to a gravimetrically determined 80% of pot capacity three times per week 26 . Subsequently, during the experimental phase, all plants were watered for four-week periods each to 80%, 60%, 50%, 40% and 30% of pot capacity, followed by re-watering at 80%. The oldest leaves of E. curvula (about a third at the beginning, reducing to zero as drought progressed) were clipped every two to three weeks to maintain the initial canopy size. For a schematic representation of the measurement and sampling strategy, see the Supporting Information in Quirk et al. 55 Leaf and root starch concentration. During the experimental phase, leaves and roots were collected from each plant every two weeks. Roots were extracted from two 2-cm-diameter soil cores taken to a depth of 8 cm and avoiding the main tap root of the trees. A total of 432 samples were microwaved, dried and ground in a mixer mill 55 . Starch concentration was analysed with a highly specific enzymatic method optimized for plant samples 73 that avoids errors of acid hydrolysis methods 74 . Starch was hydrolysed with α-amylase (Bacillus licheniformis E-BLAAM, Megazyme) and then with high-purity amyloglucosidase (Aspergillus niger E-AMGDF, Megazyme). The resulting glucose was assayed through a coupled enzymatic reaction of o-dianisidine (PGO kit, Sigma), spectrophotometrically quantified 73 . An internal reference comprising a mix of roots and leaves of all species was analysed in parallel 73 with n = 6 to quantify day drift, but the effect was not significant.
Leaf gas exchange and fluorometry. Instantaneous leaf gas exchange was measured on fully expanded leaves at midday, after ~12 photoperiod hours since watering, with an infrared gas analyser fitted with a 6 cm 2 cuvette and a red-blue LED light source (LI6400XT fitted with 6400-02B, LI-COR Biosciences). For E. curvula, three to four leaf blades were aligned, avoiding gaps and overlaps, to fill the entire leaf chamber. For the other two species, whenever leaves did not fill the chamber, leaf area was measured using scaled digital images processed in ImageJ (Fiji version 2014 for 32-bit Windows) (NIH) (Extended Data Fig. 1). To minimize environmental perturbations and error arising from CO 2 leakage from the infrared gas analyser, the leaf chamber was positioned inside the growth cabinet and supplied with air from within the cabinet at [CO 2 ] of 200, 400 or 800 ppm, depending on growth conditions. The flow rate was 235 μmol s −1 , and PPFD and temperature were set to match the growth conditions. After the readings stabilized, gas exchange was averaged for 10 s and recorded as a single point measurement, for a total of 222 data points. The same leaves were used for measurements of chlorophyll fluorescence on dark-adapted leaves (F v /F m ) at least four hours after the end of the photoperiod using a Junior PAM (Heinz Walz GmbH) and the factory settings.
Responses of net leaf A to C i and to PPFD (A-C i and A-PPFD response curves) were measured at the bench for each watering interval (n = 4, ~160 A-response curves) on the same leaves (for trees) or similar leaves (for grasses) to those used for the in-cabinet point measurements, between 3 and 72 hours after watering. For the A-PPFD curves, reference [CO 2 ] was set to 200, 400 or 800 ppm, according to experimental treatment, with ten PPFD increments between 1,500 and 0 μmol m −2 s −1 , with 5-7 min between increments. For the A-C i curves, PPFD was 1,500 μmol m −2 s −1 , and reference [CO 2 ] was incremented between 20 and 1,200 ppm, with 2-3 min between increments.
By combining these curves with empirical modelling, which does not require any assumptions 75 , the following enzyme-and light-limited photosynthetic parameters were derived: R light , Y(CO 2 ) LL , GA sat , CE, A sat , the empirical curvature of the A-PPFD and A-C i curves (m and ω, respectively) and Γ (that is, C i at which A is zero). Two parameters were derived mechanistically 75,76 : V pmax for E. curvula and V cmax for the C 3 plants. The following assumptions were made: for the enzyme-limited C 3

Plant-water relations.
We calculated E plant from ~700 gravimetric measurements. We took g s from the instantaneous gas exchange measurements described previously. We measured Ψ leaf and Ψ pd in 469 samples using a Scholander pressure chamber (PMS Instrument Company, Model 1000) on leaves cut the day or night following instantaneous gas exchange 80 . RWC was calculated for 238 samples as: RWC = 100 × sample weight−dry weight turgid weight−dry weight . Turgidity was achieved by submerging cut leaves in distilled water within a sealed environment for 3-4 h (ref. 81  Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The full dataset is available in the Supplementary Information. Source data are provided with this paper.

NaTurE PlaNTS
Extended Data Fig. 1 | experiment in progress. a) Seedlings of Celtis africana and Vachellia karroo during the growth phase before measurements began. B) A seedling of Vachellia karoo being weighed to gravimetrically determine watering amount. C) Seedlings inside the double-door growth chambers during the night phase of the photoperiod. D) Eragrostis curvula growing in the growth chambers; this picture was also taken during the dark phase of the photoperiod. The tobacco plants in C and D were grown alongside experimental plants throughout the experiment and were regularly monitored for growth and morphological traits to ensure they were reproducing treatment expected differences. e) example of in-situ leaf area measurement prior to operational gas exchange data collection. The leaf is clamped between a piece of white Perspex mounted on a white card background and it is placed at a fixed distance from the camera lens. Subsequently, images are processed in Image J to determine area as illustrated in F, wich displays monochrome images of Vachellia karroo leaflets.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection Gas exchange data were aquired through the embedded software provided by Licor, updated at the time of measurements (Open 6.3, 2015).
Images were processed with ImageJ (Fiji version 2014 for 32 bit Windows).

Data analysis
Data were treated using tools previously developed by us and that are freely downloadable (Refs 75 and 76). Statistical analyses were conducted using commercial software (SAS 9.4 and Genstat 18.2 ).
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability All figures are based on data which is made fully available in supporting information.

nature research | reporting summary
April 2020 Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
We measured at least 4 replicates per species and treatment combination. No statistical method was used to predetermine sample size. These sample size was the maximum number of plants that we could fit in the cabinets.
Data exclusions No data points were excluded from data analyses.

Replication
The effects of watering regime and CO2 levels were previously assessed separately (Refs 26 and 55) using identical setup. In this experiment CO2 levels were replicated n=2, the watering regime was applied independently to each plant three times weekly. All replications were consistent. Additionally, for control purposes, Tobacco plants ( Figure S1) whose behavior was previously characterized in detail (Ref 75) were grown alongside our experimental plants throughout the entire experiment. These Tobacco plants were regularly monitored for growth and morphological traits to ensure repeatability of the treatment. Measuring equipment was calibrated against standards and operated using best practices that we have developed previously ( Refs 73,75,76,79,80).
Randomization The plants were randomly allocated between treatments.

Blinding
The investigator was fully blinded for starch analysis and for water potential measurements: sampling was conducted independently by a second investigator and samples were identified with a code. For gas exchange and for weighing, blinding was not possible because measurements are performed in-situ on the plants and are concurrent with sampling.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.