Photosynthetic pathway and ecological adaptation explain stomatal trait diversity amongst grasses


Author for correspondence:
C. P. Osborne
Tel: +44 114 222 0146


  • The evolution of C4 photosynthesis in plants has allowed the maintenance of high CO2 assimilation rates despite lower stomatal conductances. This underpins the greater water-use efficiency in C4 species and their tendency to occupy drier, more seasonal environments than their C3 relatives.
  • The basis of interspecific variation in maximum stomatal conductance to water (gmax), as defined by stomatal density and size, was investigated in a common-environment screening experiment. Stomatal traits were measured in 28 species from seven grass lineages, and comparative methods were used to test for predicted effects of C3 and C4 photosynthesis, annual precipitation and habitat wetness on gmax.
  • Novel results were as follows: significant phylogenetic patterns exist in gmax and its determinants, stomatal size and stomatal density; C4 species consistently have lower gmax than their C3 relatives, associated with a shift towards smaller stomata at a given density. A direct relationship between gmax and precipitation was not supported. However, we confirmed associations between C4 photosynthesis and lower precipitation, and showed steeper stomatal size–density relationships and higher gmax in wetter habitats.
  • The observed relationships between stomatal patterning, photosynthetic pathway and habitat provide a clear example of the interplay between anatomical traits, physiological innovation and ecological adaptation in plants.


The stomatal pores that perforate leaf surfaces are one of the best-characterized examples of the fundamental biological relationship between form and function (Hetherington & Woodward, 2003). The area and depth of each stomatal pore, together with the density of the stomata, determine the stomatal conductance to CO2 and H2O (Brown & Escombe, 1900; Parlange & Waggoner, 1970; Franks & Beerling, 2009a; Nobel, 2009), gaseous diffusion being regulated through turgor-mediated variation in the aperture of stomatal pores (Raschke, 1975; Buckley, 2005; Franks & Farquhar, 2007). The closure of stomata under dry atmospheric or soil conditions limits CO2 diffusion from the atmosphere to chloroplasts, and means that stomatal physiology is inextricably linked to the physiology of photosynthesis (Farquhar & Sharkey, 1982). As a result, the patterning of stomata on leaf surfaces is correlated strongly with both hydrological conditions (Aasamaa et al., 2001; Sack et al., 2003; Franks et al., 2009) and photosynthetic capacity (Franks & Beerling, 2009a,b).

The evolution of the C4 pathway has caused radical increases in potential photosynthetic capacity. The C4 syndrome is one of the most important functional innovations in plants, and is particularly prevalent in grasses, where it occurs in c. 18 lineages and is utilized by around half of all modern species (Kellogg, 1999; Sage, 2004; Christin et al., 2008, 2009). The C4 pathway operates as a CO2-concentrating mechanism, elevating CO2 concentrations locally around the carbon-fixing enzyme Rubisco, with the result that the rate of its carboxylase reaction is increased (Chollett & Ogren, 1975). In combination with the saturation of Rubisco in the bundle sheath, the C4 pathway can also deplete CO2 to lower concentrations within leaf airspaces before photosynthesis is limited (Björkman, 1970; Bauwe, 1986). This, in turn, allows the same rate of photosynthesis to be maintained with a lower stomatal conductance in C4 than C3 leaves (Björkman, 1970; Long, 1999). Each evolutionary origin of C4 photosynthesis from a C3 ancestor might therefore be expected to present an opportunity for an associated reduction in the maximum stomatal conductance, providing water-use benefits over C3 sister taxa. However, this hypothesis remains untested.

Recent comparative studies of grasses have indicated that C4 photosynthesis is an adaptation to low atmospheric CO2 (Christin et al., 2008; Vicentini et al., 2008) and open habitats (Osborne & Freckleton, 2009), evolving at high temperatures and permitting the colonization of drier, more seasonal subtropical environments (Edwards & Smith, 2010). This ecological transition from forested, higher rainfall environments to drier, more open habitats is also expected to have driven the evolution of stomatal patterning and maximum stomatal conductance (Hetherington & Woodward, 2003). Grasses exhibit further distinct traits relating to the efficiency and speed of guard cell movement (Franks & Farquhar, 2007), which are also thought to have facilitated adaptation to open environments (Hetherington & Woodward, 2003). However, the extent to which the diversity of stomatal traits among grasses is linked to habitat remains unknown.

We hypothesized that, across a diversity of independent evolutionary origins, C4 grasses would consistently exhibit lower maximum stomatal conductance to H2O (gmax) than C3 grasses, associated with evolutionary shifts in stomatal patterning. Our recent work, which has emphasized the importance of controlling for phylogenetic diversity in comparisons of eco-physiological traits, has demonstrated that, on average, C4 grasses across multiple lineages operate with lower stomatal conductance than species from C3 sister lineages (Taylor et al., 2010, 2011). Here, we use comparative methods to address the following questions. Is C4 photosynthesis associated with reduced gmax compared with the C3 type? Are differences in gmax between species associated with precipitation or habitat water availability? Amongst grass lineages, do pore size and density, which determine gmax, show consistent patterns associated with photosynthetic type and ecological niche?

Materials and Methods

Species sampling and phylogeny

Species (Supporting Information Fig. S1) were sampled from C4 and closely related C3 lineages on the basis of phylogenetic information that was available in 2007 (Barker et al., 2001; Giussani et al., 2001; Aliscioni et al., 2003). Most groups included multiple species to allow for analysis within an ANOVA framework (Taylor et al., 2010). Here, we combined previously unpublished data on stomatal traits with a new phylogeny based on three plastid regions: the coding genes rbcL and ndhF, and the region encompassing trnK introns and the matK coding sequence. These markers were retrieved from GenBank and de novo sequencing was used to complete the dataset with, in most cases, the same accessions as were considered for the measurements of stomata.

Genomic DNA (gDNA) was isolated from seeds or dried plant tissues with the FastDNA Spin Kit (MP Biomedicals, Aurora, OH, USA). The three markers were then PCR amplified in multiple overlapping fragments of 600–800 bp with published and newly developed primers (Table S1). PCRs were carried out in a total volume of 50 μl, including c. 100 ng of gDNA template, 10 μl of 5 × GoTaq Reaction Buffer, 0.15 mM deoxynucleoside triphosphates (dNTPs), 0.2 μM of each primer, 2 mM of MgCl2 and 1 unit of Taq polymerase (GoTaq DNA Polymerase; Promega, Madison, WI, USA). The PCR mixtures were incubated in a thermocycler for 3 min at 94°C, followed by 37 cycles consisting of 1 min at 94°C, 30 s at 48°C and 1 min at 72°C. This was followed by 10 min at 72°C. Successful amplifications were cleaned with an Exo-SAP treatment and sequenced using Big Dye 3.1 Terminator Cycle Sequencing chemistry (Applied Biosystems, Foster City, CA, USA). All sequences were deposited in GenBank. The three markers were aligned using ClustalW (Thompson et al., 1994) and the alignments were then manually edited. The total length of the DNA markers exceeded 6000 bp per species (Table S2). A phylogenetic tree was obtained through Bayesian inference as implemented in MrBayes 3.1 (Ronquist & Huelsenbeck, 2003), under a general time-reversible substitution model with a gamma shape parameter and a proportion of invariant sites (GTR + G + I). Two different analyses, each of four parallel chains, were run for 10 000 000 generations, sampling a tree each 1000th generation after a burn-in period of 3 000 000. A consensus tree was computed on the 14 000 sampled trees (Fig. S1).

Plant material and growing conditions

Plants were raised primarily from seed. Seeds were surface sterilized before germination on water agar, and then allowed to establish in plugs of compost (John Innes Seed Compost) before transplanting into 4-l pots of topsoil (Lawnmix topsoil®; Dandy’s Topsoil, Chester, UK). A minority of species were propagated vegetatively (Arundo donax, Arundo formosana, Hakonechloa macra) and transplanted directly into pots of topsoil. Plants were grown in a heated glasshouse in Sheffield, UK, between 21st May and 18th October 2007 (daily quantum input (mol m−2 d−1): mean, 9.7; maximum, 24.7; minimum, 1.9; relative humidity (%): daily mean 64; maximum, 92; minimum, 28; temperature (°C): daily mean, 20; maximum, 28; minimum, 15; recorded using a DL2e datalogger with RHT2nl and QS2 sensors; Delta-T Devices Ltd, Cambridge, UK). Species were randomized within eight blocks and plants were watered to saturation at least twice weekly. No supplementary nutrients were provided.

Measurement of stomatal traits

The youngest fully emerged leaf was removed at the ligule from one tiller of each plant in each experimental block. Leaves were taped onto sheets of newspaper to prevent curling, and allowed to air dry in a flower press. Dental putty (President Plus-light body; Coltène/Whaledent Ltd, Burgess Hill, West Sussex, UK) impressions were taken from the mid-section of both surfaces of the preserved leaves, and nail varnish peels produced from the impressions were transferred onto Polysine microscope slides (SLS; Hessle, North Humberside, UK). Stomatal guard cell length, pore length and pore density were measured using a microscope, camera and image processing equipment (Leitz Laborlux S; Leica Quantimet 500 running Quantimet 500 Q win software, Leica Microsystems (UK) Ltd, Milton Keynes, Buckinghamshire, UK; Sanyo CCD, SANYO Sales & Marketing Europe GmbH, Watford, Hertfordshire, UK). On each slide, along a diagonal transect of the peel, five stomata were measured for guard cell and pore lengths at 400× magnification. The stomatal density on each leaf surface was determined as the mean number of stomata visible in five 0.25-mm2 fields of view, sampled along the diagonal of each peel.

Calculation of gmax

Maximum stomatal conductance to water vapour (gmax) was calculated as the sum of the maximum conductance values for each side of each leaf (g1 + g2), based on the model of Brown & Escombe (1900) after Franks & Beerling (2009a). Alternative formulations of the Brown and Escombe model have been described by Weyers & Meidner (1990: pp. 56–57) (see also discussion in Franks & Farquhar, 2001). The equation for g of one side of the leaf is

image(Eqn 1)

where the subscript i indicates the relative conductance to water vapour: = 1 for the side of the leaf with the minimum value of g and = 2 for the side of the leaf with the maximum value of g. The diffusivity of water in air (d, m2 s−1, at 25°C), the molar volume of air (v, m3 mol−1, at 25°C) and π are physical and geometric constants. The stomatal density (D, m−2) was measured as described above. The stomatal size (= guard cell length × Σ guard cell widths, m2) was calculated from our measurements of stomatal length. Following Franks & Beerling (2009a), we assumed that the depth of stomata (l, m) is equal to the guard cell width (i.e. guard cells are circular in cross-section). The maximum stomatal pore area (amax, m2) was predicted from its relationship with S, as measured from photomicrographs of fully open stomata on the leaves of 5-wk-old barley plants (grown in a glasshouse in 2-l pots of commercial compost and kept well watered). These had acclimated for several hours in full sun under water-saturated conditions. Leaf segments c. 3 cm in length were cut from mature leaves and placed directly onto the microscope stage. Within 2–3 min of excision, photomicrographs were collected using an inverted microscope equipped with a × 40 long-working-distance objective (Diaphot 200; Nikon Instruments Europe B.V., Amstelveen, the Netherlands).

Leaf level values for gmax were calculated as the sum of one-sided values for each leaf (g1 + g2). The extent to which gmax was dominated by a single side of the leaf was quantified by the ratio of the smallest to the largest of the one-sided values (g1 : g2).

Characterization of hydrological niche

The realized precipitation niche of each species was described using geo-referenced species records obtained from the Global Biodiversity Information Facility (GBIF,, accessed 26th September 2010). Species records were mapped onto 10’ grid squares defined within the Climate Research Unit CL 2.0 global climatology (New et al., 2002). Mean values for total annual precipitation, across the geographical range of each species, were calculated from precipitation values for 10’ grid cells in which each species occurred. To account for habitat-scale variation in the hydrological niche, we also compiled a list of habitats from species descriptions in regional floras (Clayton, 1970, 1989; Launert, 1971; Gibbs Russell et al., 1990; Western Australian Herbarium, 1998–; Cope, 1999, 2002; Van Oudtshoorn, 1999; Edgar & Connor, 2000; Malyschev & Peschkova, 2001; Tzvelev, 2001; Barkworth et al., 2003; Chen et al., 2006). We used these to classify species into two groups: those that were described explicitly as inhabiting wet habitats, for example, bogs, rivers, streams and water bodies (‘wet’), and those that were not (‘mesic-dry’).

Comparative methods

Analyses were carried out using species means, which were calculated from between two and eight replicates. To allow for the use of ANCOVA designs combining both discrete and continuous independent variables, we employed a phylogenetic generalized least-squares approach (PGLS, Grafen, 1989; Martins & Hansen, 1997). Correlation structures that accounted for phylogenetic covariance between species means were generated, based on pairwise shared distances on the phylogenetic tree, using Pagel’s λ (Pagel, 1999; Freckleton et al., 2002; Freckleton, 2009). Optimum values of λ were identified, and models were evaluated using a maximum likelihood modelling approach, implemented in R (Freckleton et al., 2002; pglm3.3 code available on request from R. P. Freckleton, University of Sheffield, UK). Phylogenies were edited, and the phylogenetic covariance matrix was generated using the R package ape (Paradis et al., 2004). To evaluate the robustness of predictions to the comparative method used, for those models of stomatal traits in which a simple ANOVA design was applicable, an Ornstein–Uhlenbeck (OU) approach, implemented in the R package ouch (Butler & King, 2004; King & Butler, 2009), was used to generate independent estimates of mean trait values (Table S3). Selective regimes along our phylogeny, applied in the OU models, were estimated in R via maximum likelihood using the ace function in ape, selecting the best-fitting model from symmetrical, all-rates-different and equal-rates models on the basis of the Akaike information criterion (AIC). Pagel’s correlation analysis, as implemented in Mesquite (Maddison & Maddison, 2010), was used to test for independence in the evolution of pairs of discrete traits. The likelihood test statistic was computed on the basis of 30 initial likelihood searches and 1000 simulations.


Precipitation and habitat

Precipitation niches were, on average, significantly drier for species with the C4 photosynthetic type than those with C3 (Fig. 1; Table 1, model A). This difference was not affected by the habitat occupied by each species, nor was there a significant difference in precipitation niches between species preferring mesic-dry vs wet habitats (Table 1, model A). In this model of precipitation niche as affected by photosynthetic type and habitat, the estimate of phylogenetic covariance (Pagel’s λ) was zero, that is, the modelled effects were independent of phylogenetic distance. Mean values for the precipitation niche of the C3 and C4 groups predicted by the PGLS model were highly consistent with optimum trait values obtained using the OU approach (± 6%; Table S3). The independent evolution of photosynthetic type and habitat preference along our phylogenetic tree was confirmed using Pagel’s 1994 test (difference in log-likelihoods, 0.27; P (traits independent) = 0.857). Contrasts between species of wet and mesic-dry habitats occurred within both C3 and C4 clades.

Figure 1.

Species values for precipitation niche by habitat type (wet, triangles; mesic-dry, circles) for C3 (closed symbols) and C4 (open symbols) grasses used in the screening experiment.

Table 1.   Phylogenetic generalized least-squares models used to explore differences in precipitation and habitat classification between C3 and C4 species, and the influence of photosynthetic type, precipitation and habitat classification on stomatal traits
  1. Headers indicate relevant sections in the Results section. C4, effect of C4 relative to C3; wet, effect of wet habitat relative to mesic-dry; rain, linear response to precipitation niche (mm yr−1); g1 : g2, linear response to the ratio of minimum : maximum one-sided leaf stomatal conductance.

Precipitation and habitat
 (A)loge rain = 7.12 − 0.67 C4 − 0.17 wet + 0.25 C4 wet (AIC = 40.9)
λ ≈ 0, L1 (λ0 − λ0) ≈ 0, = 1
C4 wet0.520.479
Stomatal patterning
 (B)loge D = 5.16 − 0.34 C4 (AIC = 41.7)
λ = 0.86, L1 (λ0.86 − λ0) = 10.08, = 0.002
 (C)loge S = 5.32 − 0.05 C4 (AIC = 35.4)
λ = 0.65, L1 (λ0.65 − λ0) = 6.49, = 0.011
 (D)loge S = 7.90 − 0.47 loge D − 0.71 C4 + 2.96 wet + 0.07 loge D C4 − 0.54 loge D wet − 1.62 C4 wet + 0.32 loge D C4 wet (AIC = 23.4)
λ = 0.44, L1 (λ0.44 − λ0) = 2.56, = 0.109
loge D17.27< 0.001
loge D C42.540.127
loge D wet3.080.094
C4 wet0.150.703
loge D C4 wet0.2040.656
 (E)loge S = 7.63 − 0.43 loge D − 0.34 C4 + 2.48 wet − 0.45 loge D wet (AIC = 18.9)
λ = 0.43, L1 (λ0.43 − λ0) = 3.50, = 0.061
loge D18.33< 0.001
loge D wet5.440.029
 (F)loge S = 9.09 − 0.72 loge D − 1.63 C4 + 0.26 loge D C4 (AIC = 21.2)
λ = 0.58, L1 (λ0.58 − λ0) = 7.88, = 0.005
loge D19.79< 0.001
loge D C41.160.292
Maximum leaf stomatal conductance (gmax)
 (G)loge gmax = 0.59 − 0.33 C4 (AIC = 33.2)
λ = 0.70, L1 (λ0.70 − λ0) = 7.70, = 0.006
 (H)loge gmax = − 0.42 − 0.0007 rain + 0.67 C4 + 1.73 wet − 0.0008 rain C4 − 0.0011 rain wet − 1.50 C4 wet + 0.0012 rain C4 wet (AIC = 34.8)
λ = 0.55, L1 (λ0.55 − λ0) = 4.53, = 0.033
Rain C40.1460.707
Rain wet1.660.213
C4 wet1.140.299
Rain C4 wet1.250.277
 (I)loge gmax = 0.50 − 0.32 C4 + 0.29 wet + 0.03 C4 wet (AIC = 33.1)
λ = 0.55, L1 (λ0.55 − λ0) = 4.18, = 0.041
C4 wet0.0020.964
Asymmetry between leaf surfaces (g1 : g2)
 (J)loge gmax = 0.14 + 0.89 g1 : g2 + 0.00005 C4 − 0.66 g1 : g2 C4 (AIC = 27.5)
λ = 0.81, L1 (λ0.81 − λ0) = 11.31, < 0.001
g1 : g210.260.004
g1 : g2 C42.160.154

Stomatal patterning

The allometry of individual stomata in grasses, with their distinctive dumb-bell-shaped guard cells, was derived from a variety of published photomicrographs and scale drawings (Fig. 2). The width of grass stomata is approximately equal to 0.25 × stomatal length (Fig. 2a), whereas the guard cell width, and hence the pore depth (l), is approximately equal to 0.5 × stomatal width. We found that amax was approximately 0.4 × S when measured for fully turgid barley leaves (Fig. 2b,c).

Figure 2.

(a) Stomatal size (S), as defined by guard cell length (L) and width (W). Data are values for species, based on measurements from the following: line drawings in Metcalfe (1960; triangles, apex down); photomicrographs in Flint & Moreland (1946; circle) and Kaufman et al. (1970; triangle, apex up); images of stomata from Franks & Farquhar (2007; square); and photomicrographs of barley stomata (P. J. Franks, unpublished; diamond). Dotted lines show isoclines for different values of S. Solid line shows the predicted relationship L = 3.5W + 5.0, estimated using least squares. Dashed line shows the simplified relationship, L = 4W, used for modelling purposes. (b) Relationship between pore area (a) and stomatal size (S) based on measurements from 13 images of stomata similar to (c). Solid line shows the predicted relationship = 0.33S + 47.6, estimated using least squares. Dashed line shows the simplified relationship, = 0.4S, used for modelling purposes. (c) Photomicrograph of an open stomatal pore on an attached leaf of barley (P. J. Franks, unpublished).

Because the degree of amphistomy varied between species, and some species had stomata on one side of their leaves only, we tested the effects of photosynthetic type, habitat and phylogeny on stomatal patterning by focusing on the side of the leaf that had the greatest calculated conductance capacity (g2). An initial examination of differences in S and D indicated that species belonging to the Aristida and Chloridoideae clades tended to have smaller values of S than other C4 species (Fig. 3). By contrast, six of the seven species with > 300 μm2 were members of the Paniceae tribe, and all of the species from the tribe Andropogoneae and subfamily Arundinoideae exhibited values for D that were greater than the median value for the dataset (Fig. 3).

Figure 3.

Distribution across phylogeny of the maximum stomatal conductance (gmax, 0.73–4.30 mol m−2 s−1), average size of stomata (S, 87–577 μm2), average density of stomata (D, 59–511 mm−2) and the ratio of the minimum to maximum conductance for the two sides of leaves (g1 : g2, 0–0.9). The size of the circular symbols varies in proportion to the trait values, within the ranges specified. Photosynthetic type (C3, closed symbols; C4, open symbols) and habitat (wet, triangles; mesic-dry, circles) are indicated at the tips of the phylogeny. Values of gmax are the sum of conductances for the two separate sides of the leaf.

Phylogenetic covariance in each of these stomatal patterning traits was supported by separate tests for the effects of photosynthetic type on loge D and loge S; in each case, there was evidence for a strong phylogenetic signal (Table 1, models B and C). After accounting for these phylogenetic effects, there was no significant difference in S between C3 and C4 species, but a significant 40% difference in the mean values of D (Table 1, model B) between C3 (mean, 173 mm−2; SEM, 126–238 mm−2) and C4 (mean, 124 mm−2; SEM, 94–163 mm−2) species. Although the OU method predicted a slightly larger difference in S than the PGLS method, parameter estimates for both S and D were comparable between the two methods (Table S3).

An inverse relationship, linearized by log transformation, is typically reported between S and D at the between-species level, and our data matched this expectation (Fig. 4a). After correction for phylogenetic covariance, we found no significant interaction terms in a maximal model of loge S as a function of loge D × photosynthetic type × habitat (Table 1, model D). A minimal model, produced using AIC as a criterion for the stepwise exclusion of terms, indicated that habitat preference had a significant effect on the slope of the loge S–loge D relationship (Table 1, model E), which was shallower amongst species from mesic-dry environments (Fig. 4b). The photosynthetic type had no significant effect on the slope of the loge S–loge D relationship in either model, but there were significant differences in the intercept between C3 and C4 species in both cases (Table 1, models D and E). The minimal model suggested that, for species with high D, habitat was relatively unimportant in determining S, which differed primarily between C3 and C4 species (Fig. 4b). Amongst species with low D (first quartile, 114 mm−1), habitat preference accounted for substantial differences in S: predicted S amongst C4 species from mesic-dry environments was 74% of that in wet environments, whereas, for C3 species from mesic-dry environments, predicted S was 67% of that in wet environments. Phylogenetic covariance was similar between the maximal and minimal models and did not have a significant impact on the fit of either model (Table 1, models D and E). The reduced importance of the phylogenetic covariance in these models, relative to those for the individual stomatal traits, may be a result of the strong influence of habitat on stomatal patterning. When habitat effects were not included in the initial model of loge S–loge D × photosynthetic type, accounting for phylogenetic covariance significantly improved the model (Table 1, model F).

Figure 4.

Log–log relationship between stomatal size (S) and density (D) for C3 and C4 grass species. Values are for the side of the leaf with the greatest calculated g; background shading indicates g over the range zero (black) to 3.4 mol m−2 s−1 (no shading). Wet (triangles) vs mesic-dry (circles) habitat preferences and photosynthetic types (C3, closed symbols; C4, open symbols) are highlighted. (a) Mean values for species. (b) Predicted relationships based on a minimized linear model (Akaike information criterion, AIC), accounting for the effects of phylogeny, photosynthetic type (colour scheme as in (a)) and habitat preference (solid lines, mesic-dry; dashed lines, wet) (Table 1, model E).

Maximum leaf stomatal conductance (gmax)

The tendency for C4 species to show lower D, and lower values of S for a given D on the side of the leaf with the greatest conductance value (g2), suggested that C4 species should generally have lower values of gmax (g1 + g2). A phylogenetically corrected model of loge gmax × photosynthetic type confirmed this expectation, showing a significant difference between C3 and C4 species (Table 1, model G). The model predicted that, on average, gmax was 29% lower in C4 (mean, 1.29 mol m−2 s−1; SEM, 1.06–1.58 mol m−2 s−1) than in C3 (mean, 1.80 mol m−2 s−1; SEM, 1.43–2.27 mol m−2 s−1) species. The best-fitting value of λ for this model was relatively high (0.70) and resulted in a significant improvement in model likelihood (Table 1, model G). Estimated optimum trait values for gmax in an equivalent OU model were 15% and 20% higher, respectively, for C3 and C4 species, but fell within the estimated standard error of the means based on the PGLS model (Table S3).

For the loge S–loge D relationship, we found that the phylogenetic dependence of S and D was diminished in importance when considering the effects of habitat. We therefore tested for the effects of habitat on gmax, and asked whether their inclusion in our models reduced the importance of phylogenetic covariance effects on model likelihood. When loge gmax was modelled as a function of precipitation niche, photosynthetic type and habitat, precipitation was not significant in explaining variance in gmax (Table 1, model H). By contrast, and consistent with our analysis of the loge S–loge D relationship, both photosynthetic type and habitat had significant and independent effects on gmax (Table 1, model H). The estimated value of λ for this model was lower than that for the model without habitat (0.55), but, again, provided a significant improvement in model log-likelihood (Table 1, model H). The effects of C4 photosynthesis and habitat classification on gmax were therefore detected against a background of significant phylogenetic covariance in this trait.

As the precipitation niche was strongly dependent on the photosynthetic pathway (Fig. 1), we explored the relative effects of these two factors on gmax. On the basis of the AIC criterion, no terms could be dropped from our initial model (Table 1, model H), meaning that each factor had an effect on gmax that could not be explained adequately by the other. When precipitation niche was excluded from the model, C4 species had significantly lower gmax values than C3 species, and there was an increase in the F value for photosynthetic pathway (Fig. 5; Table 1, model I). This suggests that the overall difference in gmax between photosynthetic types may be partially explained by differences in precipitation niches between C3 and C4 species. The difference in gmax attributed to photosynthetic type in the model excluding precipitation niche, remained independent of the significant difference in gmax observed between species from wet and mesic-dry habitats (Fig. 5; Table 1, model I). Optimum trait values for gmax from an OU model were consistently larger, but, again, within 20% of those predicted by the PGLS model (Table S3).

Figure 5.

Response of stomatal capacity (gmax) to photosynthetic type (closed bars, C3; open bars, C4) and habitat preference, after accounting for phylogenetic covariance. Back-transformed mean ± SEM was estimated using a phylogenetic least-squares model of loge-transformed values (Table 1, model I). Precipitation was not accounted for explicitly, but covaried with photosynthetic type, as shown in Fig. 1.

Asymmetry between leaf surfaces (g1 : g2)

The whole-leaf value of gmax comprises the sum of the predicted conductances for the two sides of the leaf (g1 + g2); therefore, the degree of amphistomy, that is, the equivalence in stomatal distribution/patterning between the sides of the leaf, measured here as g1 : g2, might be associated with gmax. If, for example, g2 is similar between species, and g1 varies, then g1 : g2 would be strongly associated with gmax. Alternatively, if increased g1 was offset by a compensatory decrease in g2, then gmax would be constant over the range of g1 : g2 from zero to unity. In the context of our comparisons, g1 : g2 might be associated with differences in gmax between photosynthetic types in two ways. First, either photosynthetic type might be more commonly associated with a specific range of g1 : g2 values. Second, if the range of g1 : g2 values is similar, an overall difference in gmax might result if the relationship between gmax and g1 : g2 differs between photosynthetic types. The median and range for g1 : g2 were similar amongst species within each photosynthetic type (C3: median, 0.52; range, 0–0.82; C4: median, 0.56; range, 0–0.91). However, although there were two species from each photosynthetic type with g1 = 0, all of the remaining C4 species (15/17, 88%) had g1 : g2 > 0.38, compared with just over one-half of the C3 species (6/11, Fig. 6). Values of gmax for species with g1 = 0 overlapped (Fig. 6) and, when loge gmax × photosynthetic type was re-tested with g1 : g2 included as a linear covariate, there was a substantial, but nonsignificant, shift in the slope of the loge gmaxg1 : g2 relationship between photosynthetic types (Table 1, model J), the slope being steeper amongst C3 than C4 grasses (Fig. 6). However, t-tests of coefficient values indicated that none of the coefficients for this model were significantly different from zero (t24 ≤ 1.53,  0.501), perhaps as a result of the uneven distribution of C4 species along the g1 : g2 axis. Although C4 photosynthesis was clearly associated with an average reduction in gmax, this analysis provides some support for the hypothesis that the difference is greatest amongst species exhibiting greater degrees of amphistomy. As with the other models of gmax presented, correction for phylogenetic covariance provided a significant improvement in the fit of the model to the data (Table 1, model J).

Figure 6.

Response of stomatal capacity (gmax) to the ratio of the minimum to maximum one-sided conductance values in C3 (closed symbols) and C4 (open symbols) grass species. Lines (solid, C3; dotted, C4) show the linear model fit after correction for phylogenetic covariance (Table 1, model I).


Our analyses support an adaptive hypothesis of stomatal evolution in grasses. First, the results indicate the correlated evolution of gmax and photosynthetic pathway. In keeping with previous work, our results also show that C4 species tend to inhabit drier precipitation niches (Edwards & Still, 2008; Edwards & Smith, 2010). However, there is little evidence that gmax is influenced by precipitation niche independently of photosynthetic type. By accounting statistically for the effects of photosynthetic pathway, precipitation niche and habitat wetness, our analyses support a relationship between stomatal traits and the physiological contrast between C3 and C4 grasses.

Overall, it was found that gmax is lower in C4 than in C3 species, mirroring the previously reported lower operating leaf conductance observed for C4 species (Taylor et al., 2010). This finding, of constitutive differences in gmax between C3 and C4 species, is consistent with well-established physiological differences between the two photosynthetic types. The role played by stomatal patterning as described by the SD trade-off, in determining this difference, is also consistent with previous studies investigating the trade-off between CO2 uptake and water loss (Hetherington & Woodward, 2003; Franks & Beerling, 2009a; Franks et al., 2009).

The physiological trade-off between carbon fixation and water loss differs dramatically between C3 and C4 species (Björkman, 1970). This trade-off has driven adaptive shifts in S and D amongst C3 species since the origins of terrestrial plants (Franks & Beerling, 2009a). Overall, S and D are negatively correlated, such that higher gmax is associated with smaller S and higher D (Fig. 4; Franks & Beerling, 2009a; Franks et al., 2009). However, reduced S and increased D can also lead to lower gmax if the reduction in S is sufficiently large, as observed for plants grown under treatment with the drought stress hormone abscisic acid (ABA) (Franks & Farquhar, 2001). The adaptation and evolution of gmax is therefore complex, and further work is necessary to elucidate the drivers and evolutionary directions of the pattern in S, D and gmax observed in this study.

The effects of the photosynthetic pathway on S and D were on the margins of statistical significance, but the phylogenetic signal was strongly supported in the model of each trait. We inferred that adaptive changes in gmax have resulted from various combinations of stomatal patterning traits, against a background of phylogenetic signals in S, D and gmax. The nonsignificant difference in the response of gmax to the degree of amphistomy observed between the photosynthetic types was also detected after correction for significant phylogenetic covariance. These results suggest constraints on the extent to which S, D and, perhaps, g1 : g2 can vary within an individual lineage, and indicate that the proximate developmental mechanisms determining gmax may depend critically on the phylogenetic group. Amongst C4 clades, for example, low gmax in Aristida species is associated with low pore density, whereas, in Chloridoideae, it is associated with small pore size (Fig. 3). Based on these differences in trait values, it seems likely that the mechanistic underpinning of differences in gmax is a further example of a similar functional outcome achieved through alternative evolutionary routes in different C4 lineages (Sinha & Kellogg, 1996; Kellogg, 1999; Christin et al., 2007, 2009).

More generally, it has been proposed that, whenever the CO2 supply becomes less limiting for photosynthesis, the high energetic costs of operating stomata should select against high D (Franks & Beerling, 2009a). In C4 leaves, physiological adaptations have reduced the limitation of CO2 uptake by CO2 supply, and our results indicate that D on the surface with the highest conductance (g2) has declined. This overall decline in D may, however, be linked with a well-characterized difference between leaves of C3 and C4 photosynthetic types: the smaller distance between vascular bundles observed in C4 species (Ueno et al., 2006), which is associated with the lower mesophyll to bundle sheath ratios diagnosing Kranz anatomy (Hattersley, 1984). As most stomata in grasses occur in rows between the vascular bundles (Metcalfe, 1960), the reduced distance between these in C4 species limits the proportion of the leaf surface area over which stomata can be distributed. It is interesting to note that, although not formally tested in this small dataset, the frequency with which C4 species showed a more even partitioning of gmax between the two sides of the leaf was higher, a phenomenon which might arise as a result of physical constraints on the development of stomata on any one leaf surface.

Our analysis suggests that there are subtle differences in effect between photosynthetic pathway and habitat in their influence on stomatal traits. Independent of the effects of photosynthetic type, we found that gmax was lower in species from dry-mesic habitats than in those from wet habitats. This is consistent with the hypothesis that stomatal patterning has evolved under selection from the degree of habitat wetness towards more or less conservative use of water. The interspecific pattern shown here, of a shallower relationship between S and D amongst species from mesic-dry habitats when compared with those from wet habitats, replicates the results of a recent intraspecific study of the impacts of water availability on Eucalyptus (Franks et al., 2009). The similarity in the outcomes of these two studies is remarkable given the potential for impacts of gross leaf morphology, for example, architectural traits associated with leaf rolling (Redmann, 1985; Heckathorn & DeLucia, 1991; Maricle et al., 2009), on stomatal patterning in comparisons of grass species from a variety of habitats.

The extent to which operational differences in leaf conductance between C3 and C4 species depend on the anatomy of stomatal patterning, as opposed to the physiological behaviour of stomatal aperture, which is considered to differ between C3 and C4 species (e.g. Jones, 1992), remains to be tested. However, our results indicate that evolutionary shifts in stomatal patterning comprise an important element in our understanding of the physiological impacts of the C4 syndrome.


We have shown that gmax, as determined by the size and density of stomata, is lower among C4 than among C3 grass species, a trend associated with a clear distinction between these photosynthetic types in terms of their precipitation niche. We have also shown that gmax is lower in grass species from mesic-dry habitats than in those from wet habitats. Our results are consistent with the hypothesis that interspecific diversity in gmax amongst grasses has arisen as a result of phylogenetic divergence in stomatal patterning, evolution of the C4 photosynthetic pathway and adaptation to habitat wetness. These results provide an excellent example of correlated evolution in physiological traits, showing how selection on physical form is mediated by physiological function.


This work was supported by the Natural Environment Research Council (NERC) standard grant NE/DO13062/1 awarded to C.P.O. and F.I.W., a Royal Society University Research Fellowship awarded to C.P.O., and National Science Foundation (NSF) grant IOS-0843231 awarded to E.J.E. P.J.F. gratefully acknowledges support from the Australian Research Council (ARC) Discovery Projects and Future Fellowships programs. We thank Guillaume Besnard for access to unpublished sequences, Elizabeth Kellogg for DNA samples, and Vernon Visser, Hui Liu and Russell Hall for critical comments on the manuscript. We thank Rob Freckleton for advice on the phylogenetic comparative methods.