Physiological advantages of C4 grasses in the field: a comparative experiment demonstrating the importance of drought

Global climate change is expected to shift regional rainfall patterns, influencing species distributions where they depend on water availability. Comparative studies have demonstrated that C4 grasses inhabit drier habitats than C3 relatives, but that both C3 and C4 photosynthesis are susceptible to drought. However, C4 plants may show advantages in hydraulic performance in dry environments. We investigated the effects of seasonal variation in water availability on leaf physiology, using a common garden experiment in the Eastern Cape of South Africa to compare 12 locally occurring grass species from C4 and C3 sister lineages. Photosynthesis was always higher in the C4 than C3 grasses across every month, but the difference was not statistically significant during the wettest months. Surprisingly, stomatal conductance was typically lower in the C3 than C4 grasses, with the peak monthly average for C3 species being similar to that of C4 leaves. In water-limited, rain-fed plots, the photosynthesis of C4 leaves was between 2.0 and 7.4 μmol m−2 s−1 higher, stomatal conductance almost double, and transpiration 60% higher than for C3 plants. Although C4 average instantaneous water-use efficiencies were higher (2.4–8.1 mmol mol−1) than C3 averages (0.7–6.8 mmol mol−1), differences were not as great as we expected and were statistically significant only as drought became established. Photosynthesis declined earlier during drought among C3 than C4 species, coincident with decreases in stomatal conductance and transpiration. Eventual decreases in photosynthesis among C4 plants were linked with declining midday leaf water potentials. However, during the same phase of drought, C3 species showed significant decreases in hydrodynamic gradients that suggested hydraulic failure. Thus, our results indicate that stomatal and hydraulic behaviour during drought enhances the differences in photosynthesis between C4 and C3 species. We suggest that these drought responses are important for understanding the advantages of C4 photosynthesis under field conditions.


Introduction
C 4 photosynthesis is a fascinating example of a complex phenotype that has evolved repeatedly (Sage et al., 2011), and influences a suite of ecophysiological traits that determine plant performance in natural settings (Long, 1999). Today, C 4 grasses are vital as agricultural crops (e.g., maize and sugarcane) and dominate the ground cover over large areas of Africa, Australia, South Asia and the Americas . The role of climate in determining the relative performance of C 4 and C 3 species from both monocot and eudicot lineages is, therefore, a key question in studies of global change (Epstein et al., 1997;Murphy & Bowman, 2006;Arnone et al., 2011;Morgan et al., 2011).
The principal physiological innovation common to C 4 lineages is the development of a biochemical CO 2 pump that operates as an extension of the dark reactions of photosynthesis (Hatch & Osmond, 1976). The C 4 pump elevates CO 2 concentrations in photosynthesising chloroplasts, virtually eliminating O 2 competition for the active site of Rubisco and therefore photorespiration, while in C 3 plants, photorespiration limits net CO 2 assimilation (A) at higher temperatures and low partial pressures of CO 2 (Osmond et al., 1982). The efficient delivery of CO 2 to Rubisco in C 4 plants improves photosynthetic efficiency at high temperatures, but bears an energetic cost that limits the maximum efficiency of photosynthesis in C 4 species at low temperatures (Ehleringer & Pearcy, 1983). The initial CO 2 assimilation step in C 4 plants, which is catalyzed by of C 3 plants. This generates the CO 2 concentrating effect of the C 4 pump and, in combination with the increased assimilation rates driven by the pump, means that C 4 leaves are able to maintain higher A at lower internal CO 2 concentrations (Collatz et al., 1992). The rates of supply of CO 2 to the intercellular spaces of leaves and the loss of water through transpiration (E) are intrinsically linked, and water use can be limited by reducing stomatal conductance (g s ; Raschke, 1975). C 4 photosynthesis therefore has important consequences for leaf water-use efficiency, i.e. net CO 2 assimilation per unit of water loss, a key observation noted in the earliest studies of C 4 ecophysiology (Black et al., 1969;Bjorkman, 1971).
Despite the important consequences of C 4 physiology for water use, until recently the primary ecophysiological explanation for C 4 grass species distributions was considered to be growing season temperature (Teeri & Stowe, 1976). For C 4 eudicots, however, adaptation to arid environments has long been accepted as important in shaping species distributions (Ehleringer & Monson, 1993;Ehleringer et al., 1997). For the grass family (Poaceae), within which the majority of C 4 species and 30% or more of C 4 evolutionary origins occur (Sage, 2009;Sage et al., 2011), habitat water availability and plant hydraulics are receiving renewed attention for their role determining the evolutionary success and distribution of C 4 species (Edwards & Still, 2008;Osborne & Freckleton, 2009;Osborne & Sack, 2012;Pau et al., 2012;Taylor et al., 2012;Griffiths et al., 2013). Recent use of molecular phylogenies has provided new insights into how evolutionary processes have shaped species distributions with respect to climate. Occupation of cooler habitats by C 3 species is now known to be associated with a preference for cooler climates in two Visser et al., 2014) of the nine monophyletic subfamilies of Poaceae (Grass Phylogeny Working Group II, 2012): Pooideae (Vigeland et al., 2013) and Danthonioideae (Humphreys & Linder, 2013), species of which are all C 3 . In contrast, comparisons between C 3 and C 4 grasses within the PACMAD clade are most appropriate to studies of the adaptive advantages of C 4 photosynthesis among grasses (Edwards et al., 2007;Edwards & Still, 2008). The PACMAD clade includes the monophyletic subfamilies Panicoideae, Arundinoideae, Chloridoideae, Micrairoideae, Aristidoideae and Danthonioideae, excludes the Pooideae, and encompasses the evolutionary origins of all contemporary C 4 grass species (Christin et al., 2009;Grass Phylogeny Working Group II, 2012). Within the PACMAD clade, the evolution of C 4 photosynthesis has resulted in preferences for drier habitats by C 4 lineages (Edwards & Still, 2008;Osborne & Freckleton, 2009;Pau et al., 2012), and divergence in water-use traits between C 3 and C 4 grasses (Taylor et al., 2012;Griffiths et al., 2013).
Paradoxically, as evidence has mounted to support the importance of drier habitats to the evolutionary success of C 4 photosynthesis in grasses, it has become clear that photosynthesis in these species may be more susceptible to failure under declining leaf water status (reviewed by Ghannoum, 2009;Driever & Kromdijk, 2013). Following restriction of watering in pot-based experiments, g S of C 3 grasses declines to a greater degree and C 3 water-use efficiency can increase to match that of C 4 plants Taylor et al., 2011). However, there is evidence that C 3 grasses commonly operate at more negative leaf water potentials (Ψ) than C 4 species Taylor et al., 2010Taylor et al., , 2011. As a consequence of these observations, it has been proposed that differences in plant hydraulics may have played an important role in allowing C 4 grasses to colonize and adapt to dry and open habitats (Osborne & Sack, 2012): decreased responsiveness of Ψ, g s and E to water availability may result in photosynthesis among C 4 grasses showing greater resistance to the effects of drought.
To date, observations of susceptibility to drought among C 4 species have been made primarily in potbased studies, which have several potential limitations (Poorter et al., 2012). There is, therefore, only limited evidence that can be used to compare the impacts of drought on the leaf physiology of closely related C 3 and C 4 species under natural growing conditions (Ripley et al., 2007;Frole, 2008;Ibrahim et al., 2008). Crucially, all of these studies have focused on comparisons within the Panicoideae subfamily, and there is no evidence addressing contrasts across other key PAC-MAD lineages. We therefore established an outdoor common garden experiment using twelve C 3 and C 4 grass species, sampled from four closely related PAC-MAD lineages. All of the species used in the experiment are found within 60 km of the study site, in a region of the Eastern Cape of South Africa where climate, according to the Koppen-Geiger classification, is warm temperate, fully humid, with warm summers (Peel et al., 2007). Our goal was to compare physiological responses of C 3 and C 4 grasses to an experimental manipulation of water availability, testing whether responses of leaf gas exchange and water potential previously observed under more controlled conditions are important under natural climatic conditions. Based on our previous experiments (Ripley et al., 2007Taylor et al., 2010Taylor et al., , 2011, we hypothesized that C 4 grasses would show higher A, lower g s, and higher water-use efficiency when well watered. During periods of progressive drought, we expected that g s in C 3 grasses would decrease to a greater extent and that differences in leaf water-use efficiency might also diminish between C 3 and C 4 grasses (Frole, 2008;Ripley et al., 2010;Taylor et al., 2011). We further hypothesized that limitation of photosynthesis observed during drought in C 3 species would be principally driven by decreased g s , but in C 4 species would instead be associated with decreased midday leaf water potential (Ψ midday ; Ghannoum et al., 2003;Ripley et al., 2007). We also predicted that C 4 grasses would show less negative Ψ midday and maintain smaller hydrodynamic gradients from soil to leaf (DΨ = Ψ predawn -Ψ midday ) when well watered, differences that we expected to be reduced under drought (Taylor et al., 2010(Taylor et al., , 2011. Finally, we aimed to test whether differences in leaf Ψ were associated with greater plant hydraulic conductance in C 4 grasses (K plant = E/ÀDΨ; Osborne & Sack, 2012).

Experimental design and plant species
Twelve grass species of open habitats were drawn from four lineages found in the regional species pool of the Eastern Cape of South Africa (Gibbs Russell et al., 1990), based on a random sample of three species per lineage (Table 1). The two C 4 groups were the genus Aristida and the tribe Andropogoneae, which share a biochemical subtype (NADP-me) but have independent origins of their C 4 syndrome (Christin et al., 2009;Grass Phylogeny Working Group II, 2012). The C 3 subfamily Danthonioideae and C 3 species from the tribe Paniceae were used in comparison; both are important components of grassland ecosystems in southern Africa.
Plants were collected from field locations (Table 1) between January 2007 and January 2008 and established in the outdoor common garden. The common garden had a blocked design, in which individual plots were separated by 2 m of short lawn, and paired 2 9 2 m plots within each of eight blocks were either watered or allowed to receive natural rainfall. Plants were regularly spaced and species locations were randomized within each plot but matched between watered and natural-rainfall plots in each block. All plots were watered on a regular basis until October 2008. After this time, only the plots in the watered treatment received additional water; approximately 28 l (equivalent to approximately 7 mm rainfall) was added to each plot every 2-3 days during the growing season. Following rainfall greater than 10 mm in 48 h, watering was halted for 2 weeks. Plots were weeded and the surrounding lawn mown on a regular basis.

Weather
Air temperature, humidity, wind-speed and direction, precipitation and photosynthetic photon flux density (PPFD) were recorded using a weather station. This comprised a datalogger (DL2e Delta T, Cambridge, UK); two relative humidity and temperature sensors (RHT2 nl, Delta T) positioned at 0.5 and

Estimation of reference crop evapotranspiration
To assess the effects of our watering treatment, R Language and Environment version 3.0.1 (R Core Team, 2013) was used to calculate reference crop evapotranspiration (ET 0 , mm day À1 ), which was compared with rainfall and watering inputs. Daily mean values (mean of maximum and minimum) from weather station measurements were used in combination with the Penman-Monteith equation, following Allen et al., (1998; Data S1). The method assumes an extensive surface of growing, green grass, completely shading the ground and not short of water. Water shortage was observed at our site and bare soil was maintained between plants, thus the calculated ET 0 is an approximate guide of true evapotranspiration.

Leaf water potential
To assess plant water deficits, Ψ midday and Ψ predawn were measured and DΨ was estimated as the difference between them, assuming Ψ predawn was equilibrated with Ψ soil . Measurements of Ψ midday were paired with measurements of gas exchange (below). For measurement, leaves were enclosed in polythene and immediately excised using a razor blade. The balancing pressure was determined using a Scholander-type pressure bomb. Ψ predawn of leaves selected using similar criteria to those used for Ψ midday was determined before sunrise within 48 h. If rainfall occurred between the collection of midday and predawn measurements, Ψ predawn measurements were either discarded or repeated the following day to better represent prevailing daytime conditions.

Leaf gas exchange
Gas exchange measurements were made during the final 2 weeks of each month during the growing season. Measurements were made under all but wet and extremely overcast conditions to obtain representative snapshots of seasonal gas exchange. During each day on which leaf gas exchange was measured, measurements were taken for one block between 09:30 h and 15:00 h. The first treatment to be measured was rotated each day, and the order of sampling between species was determined by their randomized positions within each plot. A portable open gas exchange system (LI-6400; LI-COR, Inc., Lincoln, NE, USA) was used for gas exchange measurements, equipped with a CO 2 mixer (LI-6400-01) and 30 mm 9 20 mm chamber/red-blue LED light source (LI-6400-02B). The CO 2 mole fraction of air entering the chamber was maintained at 400 lmol mol À1 . Light levels were matched to a PPFD sensor (LI-190); attached via a 1.5 m extension lead and mounted prior to measurements in each plot, in an unshaded, north-facing position, at 45°from vertical and roughly 30 cm above the soil surface. Air temperature in the chamber was not controlled, but the equipment was shaded to prevent excessive heating and to allow the chamber temperature to track that of the air. Leaf temperature was estimated using an energy balance calculation. Incoming air was not scrubbed of water vapour.
As the leaves of most species were narrow (1 to 3 mm wide), multiple leaves were usually inserted into the chamber, with a minimum of 100 mm 2 total projected leaf area used for all measurements. Leaves selected for gas exchange were the youngest fully emerged leaves on their tillers, with flowering tillers being avoided wherever possible and sections of canopy where leaf blades were exposed to full sun being preferred. Leaf area was calculated based on the known dimensions of the chamber and the combined widths of the inserted leaves at either edge of the chamber, measured using a ruler. Low fluxes were encountered regularly, especially during dry periods, forcing the use of flow rates down to 100 lmol s À1 to obtain resolvable differences in CO 2 (DCO 2 > 10 lmol mol À1 ) and H 2 O (DH 2 O > 1 mmol mol À1 ) between the reference airstream and the chamber. The chamber was tested for leaks by exhaling around the seals immediately after inserting leaves. Measurements were taken as soon as the predicted intercellular CO 2 concentration (c i ) stabilized. If c i failed to stabilize within 3 min, if DCO 2 < 10 lmol mol À1 , or if leaves being measured were thick/rolled, the chamber was re-tested for leaks and, if necessary, the seal on the chamber was readjusted before re-commencing measurements. In all cases where DCO 2 was < 10 lmol mol À1 , reference and chamber gas analyzers were matched prior to measurement.
For the first set of measurements in November 2008, rolled leaves were routinely unrolled to take measurements. Paired measurements, taken with leaves first rolled and then unrolled, indicated that by unrolling leaves, values for c i were elevated to an unusual degree due to increases in estimated g s (data not shown). Thus, from December 2008 onwards, tightly rolled leaves were not unrolled during gas exchange measurements.

Estimation of leaf transpiration
To assess water use at the leaf level, a model implemented in R Language and Environment version 3.0.1 (R Core Team, 2013) was used to estimate E for individual leaves from each species in the study. The Penman-Monteith equation (Penman, 1948;Monteith, 1965) was combined with an iterative approach to modelling of leaf energy balance for a horizontal leaf suspended over a lawn (Jones, 1992; Data S2). The model was parameterized using leaf widths based on published values for each species (Data S2), mean values for climate variables (Data S3) and g s (Data S4) from each measurement period during the growing season.

Statistical analysis
Statistical analyses were carried out using the R Language and Environment, version 3.0.1 (R Core Team, 2013). To determine the effects of the watering treatments, a Wilcoxon signed rank test was used to test for differences in weekly ET 0 À (watering+rainfall) values.
Linear mixed effect models of seasonal changes in physiological traits were fitted using maximum likelihood, and tested for significance using tools in the lme4 package (Bates et al., 2013). The data used in models were species mean values calculated for each month 9 treatment combination. Prior to analysis, mean values based on ≤ 2 replicates were eliminated from the dataset and, to improve balance in the dataset, species means that were unpaired across treatments in any given month were also removed. The full datasets used for analysis are plotted in Data S5. Average values for C 3 and C 4 groups in both treatments during each monthly sampling interval were predicted as fixed effects. Clade was modelled as a random effect dependent on the month of sampling. Model validation was carried out by inspection of residuals and, except for the model of instantaneous water-use efficiency (A/E), log-transformation was used to improve homoscedasticity of data. Bootstrapped 95% confidence intervals for fixed effect predictions were generated using 1000 simulations of each model.

Weather and effect of watering
Maximum temperatures were observed in January (mean of daily maxima during January, 28°C), whereas rainfall and relative humidity were greatest in February (total rainfall, 139 mm; relative humidity, mean of daily minima February 74%; Fig. 1a). Relative humidity was lowest during early November (mean of daily minima November 1st-15th 53%) and late March-early April (mean of daily minima March 15th-April 15th 47%; Fig. 1b).
Supplementary watering significantly reduced the cumulative water deficit, indicated by rainfall deficit, ET 0 À(watering + rainfall), on a week-by-week basis (Wilcoxon signed rank test, P < 0.001; windspeeds used to calculate ET 0 are shown in Fig. 1c). The total accumulated water deficit in the rain-fed plots was estimated to have been more than 3 times that in the plots receiving supplementary water (differences in water input are shown in Fig. 1d). Rainfall peaked during the week ending February 19th (Fig. 1d). In the 16 weeks prior to the peak of rainfall, total deficits in the watered treatments were 155 mm, compared with 386 mm in the rain-fed treatments. Peak rainfall in February was followed by a further period in which rainfall was low: rainfall deficits were 120 mm in the rain-fed plots and 15 mm in the watered plots over the final 10 weeks of the experiment. We note, however, that as an approximation of soil water balance, rainfall deficit calculated in this manner does not account for soil hydrology and depends on the method used to estimate ET 0 .

Plant survival
A number of plants died during the 2008-2009 growing season ( Table 1). The small sample size meant that there was no clear evidence that mortality for any species differed between the watered and rain-fed plots (Table 1). Compared with nine deaths in the rain-fed plots, 14 plants died in the watered plots, but six of the dead plants in watered plots were of a single species, P. ecklonii. This was one of three species for which more than two of the 16 planted individuals died; P. ecklonii (ten dead), K. curva (four dead), and P. curvifolia (four dead), are all C 3 plants. Overall, therefore, 19 C 3 plants died, compared with four C 4 plants.

Leaf water potentials
Leaf water potentials in rain-fed plots were significantly more negative than in watered plots. Watering caused significant increases in average Ψ predawn , of 0.26-1.07 MPa, for both C 3 and C 4 photosynthetic types during January and April (Fig. 2a,b). In December and March, watering led to significant improvements in average Ψ predawn for C 3 grasses (increase by 0.18-0.29 MPa), but smaller improvements for C 4 grasses (0-0.05 MPa) were not statistically significant (Fig. 2a,b).
Mean Ψ values were always more negative for the C 3 than C 4 groups (Fig. 2). Except during the wettest month, February, differences in Ψ predawn between the photosynthetic types in rain-fed plots (0.19-0.42 MPa) were statistically significant (Fig. 2a). These differences were eliminated in the watered plots, except when drought was most acute during January (0.16 MPa difference) and April (0.19 MPa difference, Fig. 2b).
Significant differences in Ψ midday (Fig. 2c,d) did not entirely mirror the pattern of drought response shown by Ψ predawn . Significant positive effects of watering on Ψ midday were observed for C 4 grasses in January (0.82 MPa) and April (0.43 MPa; Fig. 2d). However, watering significantly increased Ψ midday among C 3 plants only in January (0.79 MPa), not in April (0.24 MPa; Fig. 2d); a contrast with Ψ predawn which was affected consistently across the photosynthetic types at the two timepoints. During the wetter month of February, differences between the average Ψ midday of C 3 and C 4 grass leaves were 0.22-0.31 MPa and were significant in both rain-fed and watered plots (Fig. 2c,d): these differences were particularly notable, given the lack of differences in Ψ predawn in February (Fig. 2a,b). In addition, Ψ midday differed significantly between the photosynthetic types in the rain-fed plots during November. In the watered plots during January and April (Fig. 2c,d), significant differences in Ψ midday between the photosynthetic types were coincident with significant effects of watering on Ψ midday of one or both types (Fig. 2c,d).

Gas exchange
In each month, average A was always higher among C 4 grasses (range of means 5.1-14.7 lmol m À2 s À1 ) than among C 3 grasses (range of means 0.6-11.5 lmol m À2 s À1 ). In the rain-fed treatment, differences between the photosynthetic types during the drought periods, December-January and March-April, ranged between 3.1 and 8.2 lmol m À2 s À1 , and confidence limits indicated that they were statistically significant (Fig. 3a). In the wettest month, February, differences between the photosynthetic types were smaller in the rain-fed plots and were not significant (1.7 lmol m À2 s À1 ). This was also true at the start of the growing season in November (2.4 lmol m À2 s À1 ; Fig. 3a). During this first month of the experiment water deficits in the rain-fed plots may still have been establishing, as watering ceased during October.
In the watered plots, there were also significant differences in A between the two photosynthetic types during December, January and March (Fig. 3b), which ranged from 4.9 to 6.2 lmol m À2 s À1 . Both photosynthetic types showed significant increases in A (4.8-6.0 lmol m À2 s À1 ) in response to watering during the most severe drought periods (January and April, Fig. 3b), but only the C 3 group increased A significantly   Fig. 2 Seasonal contrasts in leaf water potential between C 3 and C 4 PACMAD grasses in a common garden experiment at Grahamstown, Eastern Cape of South Africa, between November 2008-April 2009. (a,b) Predawn water potential, Ψ predawn ; (c,d) midday water potential, Ψ midday . Points represent pooled means and 95% confidence intervals for 4-6 species in plots that were rain-fed (a,c) or given supplemental water (b,d). Values are back-transformed from the log-transformed scale used for statistical analysis. Differences between photosynthetic types and months in the experiment are highlighted by grey-filled areas. Photosynthetic type comparisons for which confidence intervals indicate significance at the P < 0.05 level are highlighted by * within each pane. Significant differences within photosynthetic types that resulted from watering are indicated by + below the relevant means in (b) and (d).
in response to watering under less severe drought during December (increase of 2.3 lmol m À2 s À1 ) and March (increase of 4.5 lmol m À2 s À1 ; Fig. 3b). Watering had no significant effect on A in the wettest month, February, or in November, at the start of the growing season (Fig. 3b).
Unexpectedly, in the natural-rainfall treatment, average g s among C 3 species was often significantly lower, by 0.031-0.056 mol m À2 s À1 , than among C 4 species (Fig. 3c); the exception was the wettest month, February, during which average g s among C 3 species was slightly, but not significantly greater than among C 4 species (0.022 mol m À2 s À1 , Fig. 3c). Differences in average g s between the photosynthetic types in the watered treatment were never significant, but showed a similar pattern to those in the rain-fed plots (Fig. 3d); the average g s of C 4 species was greater by 0.001-0.028 mol m À2 s À1 , except during February when the C 3 value was 0.038 mol m À2 s À1 greater than for C 4 species. Similar values for g s between the photosynthetic types in the watered plots were a result of significant increases in mean g s among C 3 species in December, January, March and April, relative to rainfed plots, of 0.039-0.075 mol m À2 s À1 (Fig. 3d); watering did not significantly influence the mean g s among C 4 species (Fig. 3d). These results contrasted with our expectation that well-watered C 3 plants would show significantly higher g s than their C 4 relatives.
As expected, patterns in modelled E were broadly consistent with the patterns seen for g s (Fig. 3e,f). Average values of E were 0.04-0.64 mmol m À2 s À1 higher for C 4 species when compared with C 3 species in rainfed plots, and significantly so in November, January, March and April (Fig. 3e). Watering eliminated these differences in February, and the maximum difference between photosynthetic types in the watered plots was 0.54 mmol m À2 s À1 (Fig. 3f). The smaller difference in E between C 3 and C 4 species in the watered plots resulted from watering-induced increases of 2-59% in E among C 3 species. Differences in average E in the rainfed plots were also eliminated during the wettest month, February (Fig. 3e). However, in contrast with patterns in g s , where watering had a significant influence on values in four of six months, watering had a significant effect on E among C 3 species only in the driest months, January and April (Fig. 3d,f).
To summarize, although watering always eliminated the differences in g s and E between photosynthetic types under rain-fed conditions (Fig. 3d,f), the differences in A between the photosynthetic types persisted in both watered and rain-fed treatments during drought (Fig. 3b). Thus, C 4 grasses held a photosynthetic advantage over their C 3 relatives when operating at similar E and g s . Furthermore, while decreased photosynthesis among C 3 species was associated with significant declines in g s and Ψ predawn due to drought, the same was not true for their C 4 counterparts. Among C 4 species, g s and E were not significantly affected by water supply, and decreases in A coincided instead with significant decreases in Ψ predawn . (e,f) transpiration, E. Experimental plots were rain-fed (a,c,e) or given supplemental water (b,d,f). Points represent pooled means and lines 95% confidence intervals for 4-6 species, values are back-transformed from the log-transformed scale used for statistical analysis. Differences between photosynthetic types and months in the experiment are highlighted by the grey-filled areas. Photosynthetic type comparisons for which confidence intervals indicate significance at the P < 0.05 level are highlighted by * within each pane. Significant differences within photosynthetic types that resulted from watering are indicated by + below the relevant means in (b), (d) and (f).

Water-use efficiency
Although leaf-level water-use efficiency tended to be higher, on average, among C 4 grasses, differences between the photosynthetic types during each monthly sampling interval were rarely significant (Fig. 4). At the beginning (November) and end (April) of the growing season, differences in intrinsic water-use efficiency (A/g s ) between the photosynthetic types in rain-fed plots were small (6-7 mmol mol À1 ) and were not significant (Fig. 4a). Throughout the remainder of the growing season, average A/g s of C 4 leaves in rain-fed plots was 64-125 mmol mol À1 , greater than the values for C 3 leaves (26-91 mmol mol À1 ) and significantly so during periods of intermediate drought stress in December and March (Fig. 4a). This A/g s advantage to C 4 grasses in rain-fed plots was therefore at its maximum when A, g s and Ψ predawn were water limited (i.e. significantly affected by the watering treatment) among C 3 but not C 4 grasses (Fig. 3). When drought was most severe during January, however, although the difference in A/g s between the photosynthetic types in the rain-fed plots was large (38 mmol mol À1 ) it was not significant. In contrast, during the wettest month, February, differences of a similar size to that seen in January were significant in both watered (33 mmol mol À1 ) and rainfed (37 mmol mol À1 ) plots (Fig. 4a). To summarize, advantages to the C 4 grasses in A/g s were largest in well-watered soil and during mild drought, but were lost under severe drought and were also small at the beginning and end of the growing season (Fig. 4a,b). Variation over the growing season and across treatments meant that average instantaneous water-use efficiency (A/E) ranged from 0.72 to 6.89 mmol mol À1 for C 3 and 2.29 to 7.99 mmol mol À1 for C 4 plants, and was 1.10-2.61 mmol mol À1 greater among C 4 grasses from December to March. Differences in A/E were similar to A/g s in that they usually favoured C 4 grasses and that in the rain-fed treatment they were smallest in November and April (0.14-0.5 mmol mol À1 ). Indeed, C 3 grasses showed very similar A/E to their C 4 relatives in the watered plots (À0.1-0.06 mmol mol À1 difference; Fig. 4c,d) in November and April. However, statistically significant advantages to C 4 grasses were observed for A/E in the rain-fed plots in December and March (Fig. 4c), consistent with significant differences in A/g s . We were surprised to find that, in contrast with A/g s , having accounted for the effects of leaf energy budget by calculating A/E, differences in leaf instantaneous water-use efficiency between the photosynthetic types were not significant in either treatment during February (Fig. 4c,d), the wettest month in the study.

Plant hydraulics
The size of DΨ reflects the hydraulic balance between water loss from the leaves and supply from the roots and soil, with more negative values for individual plants indicating greater strain. Average values by photosynthetic type ranged between À0.73 and À1.59 MPa. In watered plots, although differences in DΨ between C 3 and C 4 species were never significant (Fig. 5b), average values for C 4 species were consistently smaller (À0.91 to À1.38 MPa) than those for C 3 species (À1.10 to À1.59 MPa; Fig. 5a,b). In rain-fed plots, average values for C 3 species were similar to or smaller than those for C 4 plants during drought in December (C 3 À1.27 MPa, C 4 À1.29 MPa), March (C 3 À1.30 MPa, C 4 À1.51 MPa) and April (C 3 À1.04 MPa, C 4 À1.08 MPa), A/E. Points represent pooled means and 95% confidence intervals for 4-6 species in plots that were rain-fed (a,c) or given supplemental water (b,d). Values are back-transformed from the log-transformed scale used for statistical analysis (a,b), analysis was carried out on untransformed data (c,d). Differences between photosynthetic types and months in the experiment are highlighted by the grey-filled areas. Photosynthetic type comparisons for which confidence intervals indicate significance at the P < 0.05 level are highlighted by * within each pane. There were no significant differences due to watering. differences that were not statistically significant (Fig. 5a). The only significant differences in DΨ were observed under the most severe drought in January, when the C 3 DΨ (À0.73 MPa) was significantly smaller than both C 3 grasses in the watered plots (À1.39 MPa) and C 4 grasses in the rain-fed plots (À1.14 MPa; Fig. 5a). This significantly smaller value of average DΨ among C 3 grasses was observed during a period of acute leaf water deficit (Fig. 2). Therefore, the nonsignificant changes in a similar direction, which were observed in December, March and April, might also be interpreted as indicative of reduced or more variable hydraulic performance among C 3 species. A measure of whole-plant leaf-specific hydraulic conductance (K plant , mmol m À2 s À1 MPa À1 ) is provided by the flux of water due to E (mmol m À2 s À1 ) normalized by DΨ (MPa). K plant was almost always greater among C 4 species (1.32-2.69 mmol m À2 s À1 MPa À1 ) than C 3 species (1.12-2.35 mmol m À2 s À1 MPa À1 ; Fig. 5c,d), a difference of 0.2-0.88 mmol m À2 s À1 MPa À1 . The exception to the general rule that K plant was greater among C 4 species was in the rain-fed plots during January (Fig. 5c), when the difference was almost zero (Fig. 5a), coincident with the significant decline in average DΨ for C 3 species. Differences in K plant between C 3 and C 4 species were significant in November, December and April in the rain-fed plots (Fig. 5c) and in November, December, January and March in the watered plots (Fig. 5d), but watering had no significant effects on K plant (Fig. 5d).

Discussion
Our results demonstrate that plant water relations play a key role in maintaining the physiological advantages of C 4 over C 3 PACMAD grasses under field conditions. We found that A, E and g s among C 3 grasses declined significantly in response to drought, in concert with reductions in DΨ. In contrast, C 4 grasses maintained DΨ, E and g s throughout the growing season, and A was limited by water supply only under the most extreme drought conditions when both Ψ predawn and Ψ midday decreased. The findings that g s is more sensitive to drought among C 3 grasses and that A is more obviously associated with Ψ midday than with g s in C 4 grasses are consistent with evidence from previous experiments (Ghannoum et al., 2003;Ripley et al., 2010;Taylor et al., 2011). That C 4 gas exchange was relatively independent of water supply and that g s among C 3 species was commonly lower than among C 4 relatives are novel findings that highlight the importance of both taking a field-based approach and monitoring performance throughout a growing season. While our measurements of DΨ and estimates of K plant provide some support for the hypothesis that K plant is often higher among C 4 grasses, these differences were not clear-cut, and their physiological basis remains unclear. Therefore, important questions remain about the causes of C 4 resistance to drought.
We found that g s among C 4 plants was independent of our watering treatment, but A decreased in conjunction with Ψ midday under more severe drought. Among C 3 species, decreases in photosynthesis were paired with decreases in Ψ predawn and g s , but among C 4 species, decreases in Ψ predawn occurred later and g s never decreased significantly. Although they represent average responses and summarize the performance of species with sometimes distinct behaviours, these results are consistent with previous demonstrations that drought sensitivity of C 4 photosynthesis in grasses depends on metabolic rather than stomatal limitations Hydrodynamic gradient, DΨ = Ψ midday À Ψ predawn ; (c,d) hydraulic conductance, K plant = E/ÀDΨ. Points represent pooled means and 95% confidence intervals for 4-6 species exposed to natural rainfall (a,c) or given supplemental water (b,d). Values are back-transformed from the log-transformed scale used for statistical analysis. Differences between photosynthetic types and months in the experiment are highlighted by the grey-filled areas. Photosynthetic type comparisons for which confidence intervals indicate significance at the P < 0.05 level are highlighted by * within each pane. The significant difference within the C 3 photosynthetic type that resulted from watering is indicated by + below the relevant mean in (b); there were no significant differences due to watering in (d). (Ghannoum et al., 2003;Ripley et al., 2007Ripley et al., , 2010. Ultimately, photosynthesis in both C 3 and C 4 grasses was limited by drought in our experiment, but our results suggest that the cause differed and show that significant effects on C 4 photosynthesis occurred later. Consistent with our expectations, the greatest photosynthetic advantages for C 4 grasses occurred during the development of drought in the warmest parts of the growing season. We have previously shown that drought can narrow the gap in water-use efficiency between C 3 and C 4 grasses Taylor et al., 2011), which is large in well-watered, controlled conditions (e.g., Taylor et al., 2010 found that C 4 A/g s was double that of C 3 grasses). Here we show that, under natural conditions, with a relatively diverse group of PACMAD species, the intrinsic water-use efficiency advantage to C 4 species was much smaller when well watered (ca. 40% greater than C 3 ). It was often difficult to distinguish C 3 and C 4 species based on differences in A/g s , which, outside of the wettest periods in the experiment were significant only during periods of intermediate drought. Assuming that water deficits were generally greater in the field environment than controlled growth conditions, this result is consistent with our previous finding that drought reduces the difference in intrinsic water-use efficiency between C 3 and C 4 grasses (Taylor et al., 2011), but implies that advantages may be regained as water availability continues to decline. The response of A to g s is expected to saturate more quickly in C 3 than C 4 grasses (Osborne & Sack, 2012), and C 3 species under well-watered conditions often operate above the point where increasing g s results in diminishing returns for A (this study, data not shown). When faced with a need to reduce g s , the leaves of wellwatered C 3 plants initially face a relatively small penalty in A, and A/g s increases, but C 4 leaves retain a clear photosynthetic advantage at low g s (Osborne & Sack, 2012). Importantly, we found that, when we accounted for leaf energy balance, the C 4 advantage in A/g s translated into significant differences in A/E only during periods of intermediate drought.
We were surprised to find that g s was similar across the two photosynthetic types, even in watered plots. In previous comparisons of well-watered grasses from a diverse array of habitats, g s was significantly higher among C 3 grasses, though the full range of g s observed across C 3 and C 4 species overlapped substantially (Taylor et al., 2010(Taylor et al., , 2011. However, we have previously observed similar g s between C 3 and C 4 grasses in a subset of the species studied here (Frole, 2008;Ripley et al., 2010). We have also previously demonstrated that habitat water availability is important in determining stomatal trait differences among C 3 and C 4 grasses (Taylor et al., 2012), and it has been repeatedly shown that C 3 and C 4 lineages sort into distinct hydrological niches (Edwards & Still, 2008;Osborne & Freckleton, 2009;Pau et al., 2012;Visser et al., 2014). It is possible that differential sensitivity to vapour pressure deficit between C 3 and C 4 leaves (Bunce, 1983;El-Sharkawy et al., 1985) contributed to the smaller difference in g s values observed in these experiments. However, we suggest that similar g s was observed among the species in this study because they were sampled from a restricted suite of habitats within a seasonally dry climate region. It follows that the smaller differences we observed in A and A/g s may also depend on these factors. This interpretation reinforces the importance of plant water relations in structuring species assemblages, and leads to the prediction that differences in gas exchange traits between C 3 and C 4 grasses are likely to be more extreme among species from diverse habitats (Taylor et al., 2010(Taylor et al., , 2011(Taylor et al., , 2012. The clear advantage for C 4 grasses in midday gas exchange, particularly A, during the growing season implies a disadvantage to C 3 grasses that might ultimately influence their local persistence. Of the deaths observed during the 2008-2009 growing season, the majority were among C 3 plants, but they were not clearly associated with the watering treatment, a reminder that other factors may ultimately determine the local habitat preferences of these grasses (Visser et al., 2011). Seasonal differences in performance are one possibility: differences in leaf survival during winter have been demonstrated for the C 3 and C 4 subspecies of A. semialata when grown close to our field site Osborne et al., 2008), and a recent phylogenetic study investigating the grass flora of Hawaii demonstrated that the niche of C 3 PACMAD species is associated with winter precipitation (Pau et al., 2012). Seasonal differences in productivity are also important in mixed C 3 /C 4 grasslands (Ode et al., 1980;Still et al., 2003) and will no doubt be influenced by shifting patterns of precipitation and seasonality under global change. We found that performance of C 3 and C 4 grasses was most similar at the beginning and end of the growing season. A key question remaining to be tested, therefore, is whether performance and growth of our C 3 species in the late autumn, winter and early spring offset the physiological advantages of their C 4 relatives during the summer.
When the soil was wetter, leaf Ψ was less negative and DΨ smaller among the C 4 than C 3 species, consistent with our expectations (Taylor et al., 2010(Taylor et al., , 2011. During drought, we observed that DΨ decreased among C 3 grasses and was maintained among C 4 grasses. Because declines in DΨ were paired with decreasing E among C 3 species, whereas E among C 4 grasses increased with evaporative demand, we interpret the pattern of decreases in DΨ among C 3 grasses as indicating greater vulnerability of their hydraulic systems to failure under drought. Smaller DΨ was associated with more negative Ψ predawn , not less negative Ψ midday . It is plausible that night-time rehydration under severe drought was insufficient to bring Ψ predawn into equilibrium with soil Ψ in at least some of the C 3 species (predawn disequilibrium; Donovan et al., 2001Donovan et al., , 2003. Alternatively, increases of K plant , if genuine, may act to maintain E, hydration and physiological function in the face of increased evaporative demand (Jones, 1992). Changes in K plant can be regulated by several physiological processes, including changes in tissue conductance due to aquaporin activity (Kaldenhoff et al., 2008); changes in water requirements of growing tissues (Boyer, 1985); or changes in mass allocation such that water supply via roots is enhanced relative to water demand from leaf area (Maseda & Fern andez, 2006). Differences in hydraulic traits will have contributed to the overall differences in K plant in this experiment, and one realistic possibility is that less negative Ψ and smaller DΨ in C 4 grasses in this experiment was a result of better root system access to available soil water; we have previously observed that the C 4 lineages in this experiment have higher root mass ratios than the C 3 lineages (Taylor et al., 2010). However, it is important to note that, although K plant tended to be lower among C 3 species, it was not always so: although suggestive, our evidence is not sufficient to claim that a clear difference in K plant between C 3 and C 4 species was the principal driver for C 4 performance advantages. Nonetheless, our results highlight a need to address mechanistic questions about the integration of hydraulic and photosynthetic performance among grasses, ideally in field experiments and especially under drought.
The characterization of ecophysiological traits associated with C 4 photosynthesis is vital for understanding the natural diversity and ecological success of C 4 species, and the differential impacts of global change on C 3 and C 4 species. The comparisons reported here, using four lineages of PACMAD grasses sampled from the same regional species pool and grown under natural climatic conditions, are a crucial complement to previous experiments that were pot-based, carried out in controlled environments, or completed using a less diverse panel of species. Our experimental manipulation of water availability influenced contrasts in leaf physiology during a growing season and we found that C 4 photosynthetic advantages were maintained when a diverse panel of grass species were exposed to natural water shortages. Under mild drought, the C 4 advantage in A was increased, as C 3 leaves faced stomatal limitation of photosynthesis associated with earlier decreases in Ψ. We show that, under native climatic conditions in a location where both C 3 and C 4 PAC-MAD grasses are naturally abundant, water availability plays a crucial role in determining the magnitude of differences in physiological performance associated with photosynthetic type. Importantly, our experimental evidence supports a need to rigorously examine the proposition that advantages of C 4 photosynthesis in dry environments are significantly modified by dynamic responses of the stomata and hydraulic system to drought (Osborne & Sack, 2012). Understanding the interplay between C 4 photosynthesis and hydraulics will be crucial as we aim to better understand the response of plant communities to global change, including the question of why the distribution of C 4 PACMAD grasses is so strongly linked with water availability Pau et al., 2012).