2.1 Plant material and bioclimatic variables

Plants of A. semialata with various photosynthetic types (C₃, C₃–C₄ and C₄) and ploidy levels (diploid, hexaploid and dodecaploid) from 18 geographic origins (Figure 1; Table S2) were collected in this study (Bianconi et al., 2020; Lundgren et al., 2016). There were 11 populations with three replicates, four populations with two replicates, two populations with one plant, and one population with five replicates, according to the collection of plants. A total 48 individuals were involved in this study. Plants were grown in plastic pots (sized 1 L) containing John Innes No. 2 compost (John Innes Manufacturers Association) and fertilised once a month with Scotts Evergreen Lawn Food (The Scotts Company). Experiments were carried out in 2021.

The values of bioclimatic variables including annual precipitation were extracted from the WorldClim database (Alenazi et al., 2023; Fick & Hijmans, 2017) based on the geographic coordinates of collection locations of populations, using the Raster package (Alenazi et al., 2023; Hijmans, 2023) in R software (http://www.r-project.org/, version 4.2.1).

2.2 Phylogeny and genome sizes

For molecular dating, 100 sets of five randomly chosen Benchmarking Universal Single-Copy Orthologs (BUSCO) genes were selected and analysed to produce a set of trees that represented the various histories existing within the genomes. A reference-based method (Dunning, Moreno-Villena, et al., 2019; Olofsson et al., 2016, 2019), with the chromosome-scale genome of an Australian individual of A. semialata (AUS1-1; Dunning, Olofsson, et al., 2019) as a reference was used to generate consensus sequences for each BUSCO. Full details of molecular dating are provided in Raimondeau et al. (2023). A time-calibrated phylogenetic tree for A. semialata was generated based on nuclear genomes under a coalescence model by Bayesian inference as implemented in *Beast2 v.2.6.4 (Bouckaert et al., 2019; Raimondeau et al., 2023). The DNA substitution model was set to a GTR + G as its reliable across most genes (Abadi et al., 2019), and the relaxed molecular clock with a log-normal distribution of rates was applied. Each individual was set as a different species to obtain a phylogenetic tree that indicates the relationships among accessions. The specific approach of generating a summary tree and mapping the median ages on nodes of the highest credibility tree is provided in Bouckaert (2010) and Raimondeau et al. (2023).

Genome sizes of some A. semialata individuals were estimated using the Sysmex Partec Cyflow SL3 flow cytometry, following the method with minor modifications described by Doležel et al. (2007) and Bianconi et al. (2020). Ebihara with 3% PVP was applied to the fresh samples, which were analysed by a flow cytometer fitted with a 100 mW green solid-state laser (Cobalt Samba). According to the internal calibration standards, for diploid populations, Petroselinum crispum ‘Champion Moss Curled’ (4.5 pg/2 C) was used, while for populations with C-values more than three times larger than any diploids, Pisum sativum ‘Ctirad’ (9.09 pg/2 C) was used (Bianconi et al., 2020). Data for the remaining A. semialata individuals were retrieved from previous publications (Bianconi et al., 2020; Olofsson et al., 2016, 2021). Ploidy levels were assigned based on the estimates of genome size for individuals with known chromosome numbers and genome size (Bianconi et al., 2020).

2.3 Stomatal traits

Stomatal impressions were taken from two surfaces of the same leaves measured in gas exchange experiments. Dental putty (President Plus-light body; Coltène/Whaledent Ltd.) impressions were produced following the methods of Weyers and Johansen (1985) and Dunn et al. (2019), and nail polish peels made from impressions were transferred onto microscope slides. Guard cell length (GCL), guard cell width (GCW), subsidiary cell length (SCL), subsidiary cell width (SCW) and SD were measured using Image Pro-Plus 6.0 (Media Cybernetics). Guard cell surface area (GCSA) and guard cell volume (GCV) were estimated by assuming that guard cells are shaped like capsules, that is, cylindrical with hemispherical ends (Raven, 2014). Full details are provided by Raven (2014). GCSA to volume ratio (GCSA:GCV) was calculated as GCSV/GCV.

2.4 Leaf gas exchange

All photosynthetic parameters were measured using the LI-COR 6400XT portable gas-exchange system with a 6400-40 fluorometer head unit (LI-COR). Gas exchange measurements were carried out on the youngest fully expanded leaf of well-watered plants. The LI-COR cuvette conditions during gas exchange measurements were controlled as follows: a CO₂ concentration of 400 µmol mol⁻¹, a leaf temperature of 24 ±  1°C, a leaf vapour pressure deficit in the range 1.4–1.9 kPa, and a flow rate of 500 μmol s⁻¹.

The responses of net CO₂ assimilation (A) and stomatal conductance (g_s) to a step-change in photosynthetic photon flux density (PPFD) were carried out as described using the method of McAusland et al. (2016), with the following changes to the protocol: (i) leaves were equilibrated to a PPFD of 100 μmol m⁻² s⁻¹ until both A and g_s were at a steady state; (ii) PPFD was then increased to 1000 μmol m⁻² s⁻¹ for 1 h until a new steady-state was reached with A and g_s recorded every 30 s, before returning to 100 μmol m⁻² s⁻¹ for a further 1 h to reach the final steady state.

The same leaf was used during the same day to record the relationship between steady-state g_s and irradiance. Measurements were recorded every 30 s starting from low irradiance to high irradiance levels (0, 50, 100, 200, 400, 800, 1200, 1600, 2000 μmol m⁻² s⁻¹). Each irradiance level was provided for at least 30 min to reach a steady-state value of g_s (Ng & Jarvis, 1980).

2.5 Modelling stomatal responses to a fluctuating environment

2.6 Statistical analysis

R software (http://www.r-project.org/, version 4.2.1) was used to statistically analyse the data and generate graphs. Tests of significant differences were identified using a non-parametric test (specifically the Kruskal–Wallis test) to determine the significance of the differences in means among groups (photosynthetic types plus ploidy levels). Phylogenetic generalised least squares (PGLS) models were performed to test for correlations while considering evolutionary relationships. Values of Pagel's lambda from zero to one represented the strength of the phylogenetic signal. A summary function in R was then run for each model to determine the adjusted multiple correlation coefficients (r²) and statistical significances of covariates between variables (p < 0.05).

3.1 Stomatal characteristics

The non-parametric Kruskal–Wallis test showed that C₄ diploid grasses had significant shorter (p < 0.05) GCLs than either non-C₄ diploid populations or C₄ hexaploid and dodecaploid populations (Figure 2a). The analysis of ancestral state reconstructions for GCL in all diploids also showed that the common ancestor of C₄ A. semialata had shorter GCLs than those of C₃–C₄ or C₃ plants (Figure 2a; Figure 3 node Q vs. nodes W and Z), indicating that cell shortening coincided with the origin of C₄ photosynthesis. However, there were no significant differences in GCW, GCV, GCSA to volume ratio (GCSA:GCV) and SD among diploids (Figure 2). GCL and GCW became larger as the ploidy levels increased in C₄ plants (Figures 2 and 3). Specifically, among C₄ grasses, there were significant positive associations (p < 0.05) between genome size and guard cell size (Figure 4a,b), which is consistent with previous studies (Beaulieu et al., 2008; Jordan et al., 2015; Lomax et al., 2009). However, for a given genome size, C₃ and C₃–C₄ plants had longer guard cells with a similar width to C₄ plants (Figures 2 and 3). GCL was therefore similar, but GCW was greater, in C₄ hexaploids compared with C₃ and C₃–C₄ diploids (Figure 2a,b). As a consequence, GCV was lowest in C₄ diploids, followed by C₃–C₄ and C₃ diploids, C₄ hexaploids, and C₄ dodecaploids (Figure 2c). However, C₃ and C₄ diploids had a significantly greater (p < 0.05) ratio of GCSA:GCV than hexaploids and dodecaploids (Figure 2d).

SCLs and widths in C₄ diploids were both significantly lower (p < 0.05) than in C₃–C₄ diploids, C₄ hexaploids and dodecaploids (Figure 2e,f). SD in C₄ and non-C₄ diploids was significantly higher (p < 0.05) than that in dodecaploids (Figure 2g). However, there were no significant differences in SD among the diploids with different photosynthetic types (Figure 2g). Calculated g_max based on stomatal size and density was significantly lower (p < 0.05) in C₄ populations than in C₃ and C₃–C₄ populations (Figure 2h). Compared to C₃ diploids, the g_max value was reduced by 40% in C₄ dodecaploids (Figure 2h). There was an offset in the log-log relationships between guard cell size and SD for C₄ plants in comparison with non-C₄ plants, particularly when guard cell size was measured via cell length (Figure 4c). However, the equivalent offset for the relationship with GCW was still able to distinguish C₃ from the combination of C₃–C₄ and C₄ types (Figure 4d).

3.2 Dynamic responses of photosynthesis and stomatal conductance to light

C₃ and C₄ individuals showed different stomatal responses to a step change in the light environment, especially in dynamic responses of g_s (Figures 5 and S2–S4). For the dynamic responses of photosynthetic rate, the initial value of net CO₂ assimilation rate (A₀) at the initial PPFD of 100 µmol m⁻² s⁻¹ showed no significant variation, such that A₀ was lower than 5 µmol m⁻² s⁻¹ for all groups (Figure 5a). The rapid rise in net leaf CO₂ assimilation rate (A) in response to an increase in PPFD to 1000 µmol m⁻² s⁻¹ occurred within the first 10 min in all groups. However, it took double the time for A to reach a maximum steady state (A_s) in the C₄ dodecaploids than in the others. As expected, C₄ plants had higher A_s under high PPFD than C₃ and C₃–C₄ intermediates plants (Figure 5a).

The stomatal opening phase was characterised by an increase in g_s to a maximum steady-state value (G_s). The maximum rate of stomatal opening (Sl_{max_i}) of C₃ plants was the steepest, followed by C₃–C₄ intermediate plants, C₄ diploid plants, C₄ hexaploid plants, and C₄ dodecaploid plants (Figure 5b; Table S3). In particular, the maximum slopes of C₄ hexaploids and dodecaploids were significantly (p < 0.05) lower than those of C₃ diploids (Table S3). The rates of stomatal opening also differed among C₄ plants with differing ploidy levels (Figure 5b; Table S3). As expected, C₄ diploids had a significant (p < 0.05) higher Sl_{max_i} than C₄ dodecaploids (Table S3). Fitted model parameters (for details, see Figure S1) describing the stomatal opening response showed significant differences (p < 0.05) in non-C₄ plants compared to C₄ dodecaploid plants (Table S3). In particular, C₃ and C₃–C₄ diploids had significantly lower (p < 0.05) values of the opening time constant (k_i), representing faster stomatal opening than dodecaploids (Table S3). To isolate the effects of photosynthetic pathway evolution from those of polyploidy, which only occurs in the C₄ types, we made an ancestral state reconstruction of k_i for the diploids only. This showed that the common ancestor of C₄ A. semialata had a similar stomatal opening speed to that of C₃–C₄ plants (Figure 6 Node q vs. Node w; Table S3). In contrast, the common ancestor of C₄ and C₃–C₄ A. semialata had a slower stomatal opening speed than that of C₃ plants (Figure 6 Node p vs. Node z; Table S3).

On the other hand, stomatal closure in response to a decrease in PPFD from 1000 to 100 µmol m⁻² s⁻¹ followed a similar pattern in all groups (Figures 5 and S5), despite C₄ dodecaploids displaying the lowest value of final steady-state A and g_s (Figure 5). Compared to the speed of stomatal responses to increasing light conditions, the speed of stomatal closing was substantially faster. Nevertheless, its rapidity was still an order of magnitude slower than that of A. An almost immediate decrease of A to a new target of steady-state was established before g_s reaching the final steady-state value.

3.3 Factors explaining variation in stomatal opening speed

In general, PGLS analysis showed relationships between cell size and parameters of stomatal opening responses in C₄ plants (Figure 7). There was a significant (p < 0.05) positive relationship between the opening time constant (k_i) and GCL, SCL or GCV across the C₄ plants (Figure 7). Larger guard cells, subsidiary cells, and higher GCV were characterised by higher values of k_i, corresponding to slower stomatal opening speed (Figure 7). For a given GCL, C₄ hexaploid plants had a slower stomatal opening speed than C₃ diploid plants (Figure 7a).

Besides the effects of guard cell size and photosynthetic pathway, local climates played an additional role in stomatal behaviour. Figures 6 and 8a showed significant variation (p < 0.05) among C₄ diploids in the values of k_i, which ranged fourfold from 1.7 to 6.8 min. To isolate the effects of environment from those of photosynthetic type and polyploidy, we therefore analysed the relationship of k_i in C₄ diploids to climate. Crucially, there was a significant (p < 0.05) negative relationship between k_i and annual precipitation among the seven populations of C₄ diploid plants (Figure 8a). This means that the speed of stomatal opening was faster in C₄ diploids that have migrated into wetter environments. The opening lag time (λ_i) for stomatal opening was also significantly (p < 0.05) influenced by annual precipitation (Figure 8b). However, unlike the rapidity of stomatal opening, C₄ diploids in the wetter environments had longer lag times in their responses to increasing light (Figure 8b).

Finally, regressions of the light-saturated photosynthetic rate (A_sat) versus opening time constant k_i for A. semialata in C₄ type showed that populations with higher values of A_sat had a faster stomatal opening speed (Figure 9). In addition, C₄ plants had slower opening stomata (higher k_i) than C₃ plants for a given photosynthetic rate (Figure 9).

4.1 Guard cells shortened following the evolution of C₄ photosynthesis in diploids

The common ancestor of C₄ A. semialata had shorter guard cells than C₃–C₄ or C₃ plants (Figures 2a and 3), implying that the initial evolution of C₄ photosynthesis led to a shortening of guard cells before diversification in this species. This finding, in conjunction with previous work showing that smaller guard cells for a given density is a general property of C₄ grass stomata (Taylor et al., 2012), suggests that the size of guard cells is adjusted after the evolution of C₄ photosynthesis, and indicates a close link between photosynthetic physiology and guard cell size. One of the potential explanations for this would be that the rapid light-induction of C₄ photosynthesis requires faster stomatal opening, which is facilitated by smaller guard cells.

There is no coincident change in GCW during C₄ evolution in A. semialata (Figures 2b and S6). Width changes after polyploid formation in C₄ A. semialata show that these are possible if all cell types across the leaf are enlarged. However, within a ploidy level, width may be limited in these grasses by the development of stomatal complexes within ordered rows of cells (i.e., cell files), such that their proximity to adjacent rows constrains their width (McKown & Bergmann, 2020). GCW may be further limited by association with subsidiary cells, whose shape and signalling responses are highly coordinated with those of guard cells (Gray et al., 2020; McKown & Bergmann, 2018). However, SCW did decrease after the evolution of C₄ photosynthesis (Figure 2f), showing that width changes are possible within stomatal complexes. This may be important, because subsidiary cells are thought to adjust the range of pore apertures and stomatal responsiveness (Raissig et al., 2017). In fact, subsidiary cells may be important regulators of the speeds of stomatal opening and closure in grasses (Chen et al., 2017; Raissig et al., 2017).

SD, on the other hand, shows considerable overlap among diploid populations of A. semialata. Even though the C₃ populations have significantly higher SD than C₄ hexaploid and dodecaploid populations (Figure 2g), two of the C₃ populations have values that overlap with the C₄ range (Figure 4c), showing that this trait is not directly associated with photosynthetic type in the same way as guard cell size (Figures S7–S9). More generally, variation among other species in stomatal size and density is inversely related, such that C₄ species tend to have lower stomatal densities than C₃ plants (McAusland et al., 2016; Taylor et al., 2012). Our data imply that this general difference in density arises as a secondary adaptation after the change in guard cell size.

Among the C₄ populations of A. semialata that differ in ploidy levels, larger guard cells have a slower stomatal response to increasing light, which is consistent with previous studies of cell size effects on stomatal dynamics (Drake et al., 2013; Franks & Beerling, 2009; Hetherington & Woodward, 2003; Lawson & Blatt, 2014). The size relationship arises from a lower surface-area-to-volume ratio in larger guard cells, which limits the capacity of ion pumping and solute transport across the plasma membrane to increase cellular concentrations of osmoticants (Lawson & Blatt, 2014; Raven, 2014). Polyploid formation and its association with larger cells therefore leads directly to slower stomata. However, cell size is not the only determinant of stomatal opening and closing speeds, and other factors including guard cell shape, mechanical constraints and biochemical characteristics also play critical roles (Franks & Farquhar, 2007; McAusland et al., 2016; Ozeki et al., 2022; Raven, 2014).

If all else were equal, the evolution of shorter guard cells in C₄ diploid populations of A. semialata would therefore be expected to increase both opening and closing speeds in response to fluctuating light. However, our data indicate that this is not the case, implicating further changes in stomatal biology during the transition to C₄ photosynthesis.

4.2 Transition to C₃–C₄ intermediate physiology initially slows stomatal opening

Our data show that some of the C₄ diploids have a similar stomatal opening speed to those of C₃ and C₃–C₄ plants, such that there is no overall difference in speed among diploids with different photosynthetic types (Figure 6; Table S3). However, there is considerable variation in opening speed among the diploid C₄ plants, despite limited variation in GCL or GCV (Figures 2a,d and 6). The evidence therefore shows a marked diversification of stomatal function in A. semialata following the evolution of C₄ photosynthesis. But did the first C₃–C₄ or C₄ plants in this species have stomata that opened at a slower or similar speed to those of their C₃ sisters?

Ancestral state reconstruction of the opening time constant among diploids shows that the common ancestor of C₄ and C₃–C₄ plants in our study had a slower opening speed than that of the C₃ plants (Figures 6 and S10). Subsequent diversification within the C₄ clade has been associated with mean annual rainfall (Figure 8). Ancestral state reconstruction shows that C₄ physiology emerged once in A. semialata in a relatively dry climate (Figure 8). In summary, this evidence therefore indicates that the initial divergence between populations using C₃ photosynthesis, from those in which carbon fixation occurred partially (C₃–C₄) or wholly (C₄) via PEPC, was associated with a change in the stomatal opening speed (Figure 6). Subsequent diversification among C₄ populations as they migrated across the paleo-tropics led to the repeated evolution of faster stomata in wetter climate regions (Figure 8a).

A potential mechanism underlying these changes in stomatal opening rate is implied by the relationship of k_i to A_sat. The best current models assume that stomatal opening speed is biochemically limited by the ATP supply from cyclic and noncyclic photosynthetic electron transport (Lawson & Blatt, 2014; Matthews et al., 2020; Ng & Jarvis, 1980; Raven, 2014). Bellasio et al. (2017) developed a dynamic hydro-mechanical and biochemical model which showed that a lag in ATP production may mechanistically account for both the lag phase of stomatal response to increasing light, and stomatal opening speed. A lowering of photosynthetic capacity and/or disruption of guard cell ATP supply during the emergence of C₃–C₄ and C₄ photosynthesis are therefore plausible explanations for our observation of slower stomatal opening. The observation that the initial, maximum rates of stomatal opening are similar in C₄ and non-C₄ plants (Figures 5, S5, and S11; Table S3) support this explanation, implying that opening is limited by the supply of energy or solutes, rather than the ion pumping capacity of guard cells. An alternative explanation could be that the blue or red light sensitivity of stomatal opening are disrupted by C₄ pathway evolution. However, new evidence suggests that these mechanisms are conserved across C₃ and C₄ types of A. semialata (Bernardo et al., 2023).

C₄ photosynthesis is unlikely to be very efficient when it first evolves. In A. semialata, the process arises through a small number of changes to anatomy and biochemistry (Dunning, Moreno-Villena, et al., 2019; Lundgren et al., 2019). Lateral gene transfer subsequently contributes to more rapid improvements in the catalytic efficiency of key C₄ enzymes, including PEPC (Phansopa et al., 2020). At the same time, leaf gas exchange is also likely optimised through the regulation of SD, and the formation and patterning of mesophyll airspace (Lundgren et al., 2019; Sugano et al., 2010). Our data imply that biochemical integration of stomatal function with C₄ photosynthetic metabolism may be an important aspect of that optimisation process.

4.3 Adaptation to wetter climates accelerates stomatal opening speed in C₄ plants

Guard cell size and photosynthetic type are the dominant effects on stomatal opening speed in A. semialata. However, within the C₄ diploid type, the speed of stomatal responses to increasing light conditions differs significantly among populations (Figures 6 and 7), implying secondary adaptation of stomatal opening speed during diversification (Drake et al., 2013; Raven, 2014). This observation could be explained partly in the context of environmental characteristics across the geographical distribution of this species, with the stomatal opening time constant significantly related to annual precipitation in C₄ diploids (Figure 8a). This shows that when A. semialata C₄ plants have migrated into wetter habitats, their stomatal opening adapted by becoming faster. However, we also observed that the opening lag time was also significantly correlated with annual precipitation (Figure 8b), such that a longer lag time is coupled to a faster opening speed when C₄ diploids migrate into wetter environments.

Xiong et al. (2018) showed that a terrestrial herb (Gossypium hirsutum) had slower stomatal opening than a wet habitat species (Centella asiatica). Slow stomatal opening during shade-sun transitions may transiently limit CO₂ assimilation, if water is sufficient. The observed change in opening speed may serve as an adaptation to overcome this limitation in environmental contexts where water conservation is not critical. Conversely, the water-conserving effects of stomata that are not fully open may be more important for plant performance in drier habitats than transient CO₂-limitation.