Morphological and anatomical determinants of mesophyll conductance in wild relatives of tomato (Solanum sect. Lycopersicon, sect. Lycopersicoides; Solanaceae)

Authors


Abstract

Natural selection on photosynthetic performance is a primary factor determining leaf phenotypes. The complex CO2 diffusion path from substomatal cavities to the chloroplasts – the mesophyll conductance (gm) – limits photosynthetic rate in many species and hence shapes variation in leaf morphology and anatomy. Among sclerophyllous and succulent taxa, structural investment in leaves, measured as the leaf dry mass per area (LMA), has been implicated in decreased gm. However, in herbaceous taxa with high gm, it is less certain how LMA impacts CO2 diffusion and whether it significantly affects photosynthetic performance. We addressed these questions in the context of understanding the ecophysiological significance of leaf trait variation in wild tomatoes, a closely related group of herbaceous perennials. Although gm was high in wild tomatoes, variation in gm significantly affected photosynthesis. Even in these tender-leaved herbaceous species, greater LMA led to reduced gm. This relationship between gm and LMA is partially mediated by cell packing and leaf thickness, although amphistomy (equal distribution of stomata on both sides of the leaf) mitigates the effect of leaf thickness. Understanding the costs of increased LMA will inform future work on the adaptive significance of leaf trait variation across ecological gradients in wild tomatoes and other systems.

Introduction

Leaves are the primary photosynthetic organs in most plants and many disparate features of leaf morphology and anatomy, including cell shape, stomatal patterning and chloroplast distribution, can affect photosynthetic efficiency. Photosynthetic performance also affects components of plant fitness (Arntz & Delph 2001), so that, unsurprisingly, many features of leaves are shaped by natural selection (Givnish 1987; Smith et al. 1997; Nicotra et al. 2011). The effect of leaf structure on fundamental processes such as CO2 diffusion and hence photosynthetic performance has potentially important consequences for ecologically important differences between closely related species and for phenotypic plasticity within species. To determine what aspects of trait variation are most critical at the ecological and microevolutionary scales where natural selection acts, studies must examine variation in these traits and their relationships among diverse yet recently diverged species.

Different aspects of physiological response and photosynthetic performance are more or less determined by leaf structural variation. In particular, unlike stomatal conductance (gs), which responds physiologically within minutes to changes in the external environment, the mesophyll conductance (gm: the conductance of CO2 from the substomatal cavity to the sites of carboxylation in the chloroplast) reflects, to a large degree, structural parameters such as leaf thickness, cell packing, shape and wall thickness (Terashima et al. 2011; Tosens et al. 2012b; Tomás et al. 2013), although rapid changes in gm have been documented (Flexas et al. 2008; Warren 2008; Evans et al. 2009; Tazoe et al. 2011; Griffiths & Helliker 2013). These structures cannot be altered until new leaves develop (plastic responses) or until natural selection alters leaf structure following a changing environment (evolutionary responses). In addition, gm is increasingly recognized as a significant limitation on photosynthetic rate in many species (Flexas et al. 2008, 2012) that affects carbon-, water- and nitrogen-use efficiencies (Barbour et al. 2010; Buckley & Warren 2013; Flexas et al. 2013). Understanding the determinants of mesophyll conductance is an active area of theoretical and empirical investigation (Tholen & Zhu 2011; Sharkey 2012; Tholen et al. 2012), but little is known about their potentially important evolutionary role in shaping leaf phenotypes in response to selection across environmental gradients.

Despite the variety of leaf morphology and anatomy observed among species and environments, some aspects of leaf structure are constrained by the fundamental physics of gas diffusion (Parkhurst 1994). As CO2 diffuses into a leaf, it must pass though the boundary layer, the stomatal pore, the air-filled pore space within the leaf, the cell wall, plasma membrane and cytoplasm as well as the chloroplast membrane. Variation in leaf structures influences the conductance to CO2 at each step of this diffusion pathway (e.g. Schuepp 1993; Flexas et al. 2006; Terashima et al. 2006), and therefore can impact photosynthetic performance. Leaf dry mass per area (LMA) is one of the many aspects of leaf structure that affect gm, and is therefore one such phenotype that could be shaped by selection on photosynthetic performance. LMA is a composite of many underlying traits, such as mesophyll thickness, cell wall thickness, cell shape, cell packing and so forth. At a basic level, LMA can be decomposed into bulk leaf thickness and density (Witkowski & Lamont 1991; Niinemets 1999). A thicker leaf allows greater photosynthetic rate under high irradiance, explaining the strong plasticity of this trait in response to light (Poorter et al. 2010). Increased leaf thickness is also associated with succulence, which can prevent cellular dehydration by buffering the transpirational stream (Ogburn & Edwards 2010). Similarly, dense, sclerophyllous leaves are associated with cellular drought tolerance (Niinemets 2001, but see Bartlett et al. 2012) and/or mechanical defences against herbivores (Turner 1994; Onoda et al. 2011). Variation in LMA is therefore ecologically and evolutionary significant (Reich et al. 2003; Wright et al. 2005; Poorter et al. 2009). Generally, higher LMA is associated with greater (a)biotic stress tolerance but slower growth. Low LMA, indicating relatively little investment per leaf area, benefits plants by capturing light more efficiently and allowing faster whole plant growth and competitive ability (Poorter & Bongers 2006), often at the expense of stress tolerance. LMA is known to be highly variable between species (Reich et al. 2003; Wright et al. 2005; Poorter et al. 2009), and the optimal LMA for a given ecological strategy and/or environmental context depends on balancing these costs and benefits.

Studies across broad functional groups and within species indicate that high LMA limits photosynthetic efficiency by reducing the mesophyll conductance (Flexas et al. 2008; Niinemets et al. 2009a; Galmés et al. 2011). This is particularly evident among sclerophylls (Niinemets et al. 2009b; Hassiotou et al. 2010) and succulents (Maxwell, von Caemmerer & Evans, 1997), where extreme differences in anatomy among distantly related species explain variation in mesophyll conductance. In comparison, little is known about how LMA affects mesophyll conductance among species on the low LMA end of the spectrum (e.g. between herbaceous species) or whether LMA-mediated changes in mesophyll conductance are important over short evolutionary timescales (within species or between recently diverged species). A meta-analysis of LMA and gm showed that while LMA sets up an upper limit on gm, wide variation in gm is possible on the low LMA (herbaceous) end of the spectrum (Flexas et al. 2008), suggesting that the relationship between LMA and photosynthetic performance in herbaceous species might be highly variable from one taxonomic group to another (Hanba et al. 1999).

Predicting the relationship between LMA and gm in herbaceous leaves is difficult because the two basic components of LMA (leaf thickness and density) affect gm in different ways. Firstly, in terms of leaf thickness, thicker mesophyll in tender, herbaceous leaves might actually increase gm by allowing more chlorophyll surface area per leaf area (Sc), which provides more parallel paths for CO2 diffusion from the intercellular airspace (IAS) into the chloroplasts (Terashima et al. 2006; Galmés et al. 2013). However, greater airspace resistance in thicker leaves eventually limits diffusion (Parkhurst 1994), especially in hypostomatous species (Syvertsen et al. 1995). Consequently, the relationship between leaf thickness and gm could be positive, negative or even quadratic depending on the domain of leaf thickness over which gm is measured. Furthermore, the ratio of stomatal density on the adaxial (‘upper’) to abaxial (‘lower’) surface, referred to as the stomatal ratio (SR), alters the effective leaf thickness. In terms of CO2 diffusion through the IAS, amphistomatous leaves are essentially half as thick as comparable hypostomatous leaves, further complicating the relationship between thickness per se and gm.

The relationship between LMA and gm can also be affected traits that affect leaf density rather than thickness since increased cell packing increases airspace resistance by reducing the porosity of the mesophyll (Nobel 2009), usually measured as IAS fraction of leaf volume (fIAS). However, in herbaceous species, leaves are generally not densely packed, so these factors might be relatively less important. Leaf density is also affected by mesophyll cell walls and chloroplast envelope thickness, which are the major components of liquid phase CO2 diffusion. Because CO2 diffusion through water and lipids is many orders of magnitude slower than through air, the liquid phase resistance explains much of variation in total gm across species (Evans et al. 2009). Because of these potentially complex contributions of anatomical leaf traits to CO2 diffusion, the relationship between variation in LMA and gm in herbaceous species is currently difficult to predict in any particular group.

Wild tomato species provide an excellent opportunity to examine questions of the functional significance of physiological and anatomical traits to photosynthetic performance, over recent evolutionary history. The clade is young [<2.7 million years old (Kamenetzky et al. 2010)] but ecological niches span deserts, tropical rain forests and highlands (>4000 m) and leaf morphology is correspondingly diverse (Moyle 2008; Peralta et al. 2008). Ecological niche modelling indicates that climate variables thought to affect leaf morphology (temperature and precipitation) are the strongest contributors to niche divergence (Nakazato et al. 2010). However, few studies have examined the functional and/or ecological consequences of leaf trait variation in these species (Nakazato et al. 2008, 2012; Chitwood et al. 2012). Examining how variation in leaf structure influences photosynthetic performance within and between tomato species therefore could provide insight into the selective pressures governing the evolution leaf morphology and anatomy as well as identify those leaf traits most likely to contribute to plasticity and local adaptation. For example, to contribute significantly to ongoing evolutionary responses, there must be genetic variation in a trait within or among species. Therefore, even if one or more components of the CO2 diffusion pathway accounts for a large fraction of overall photosynthetic limitations, if this trait does not vary, it will be less evolutionarily significant than a trait that remains functionally variable within and between species.

To ascertain the evolutionary significance of leaf morphological and anatomical variation on gm and photosynthetic performance, we measured anatomical and physiological variation in eight phenotypically diverse species of closely related wild tomatoes, in order to address four questions:

  1. Do physiological traits, including gm, vary significantly within and/or between wild tomato species?
  2. How much does gm limit photosynthetic rate?
  3. How does LMA affect gm?
  4. What are the anatomical determinants of the LMA–gm relationship in these species? Specifically, how do leaf density, thickness and/or Sc determine gm?

Materials and Methods

Plant material and cultivation

We obtained seeds of eight wild species (Supporting Information Table S1) from the Tomato Genetics Resource Center (TGRC) at UC Davis (http://tgrc.ucdavis.edu). On 19–20 August 2010, seeds were soaked in 50% household bleach for 30 min, rinsed thoroughly and placed on moist paper in plastic boxes and germinated in a growth chamber. After 1 week, seedlings were transplanted to cell-pack flats containing Metro-Mix 360 (Sun Gro Horticulture, Vancouver, British Columbia, Canada). Two weeks later, seedlings were transplanted again to 3.78 L pots containing a 1:1 mixture of compost soil and Metro-Mix 360. Plants were positioned randomly in the Indiana University greenhouse and grown under supplemental lighting to maintain a constant 16:8 h light:dark cycle. Plants were irrigated to field capacity daily to prevent drought stress and fertilized weekly with a nitrogen, phosphorus and potassium solution. We periodically trimmed plants and removed any developing fruit to maintain rapid vegetative growth throughout the experiment. Gas exchange measurements and tissue for anatomical and morphological measurements were taken from individual plants in a haphazard order between November 2010 and February 2011. All traits measured are listed in Table 1.

Table 1. Partitioning phenotypic variance within and between species. Most of the variation in physiological traits (upper half) is within species. In contrast, interspecific differences account for a large fraction of the total variation in anatomical traits (lower half). Variance components were estimated using restricted maximum likelihood (REML)
TraitVariance component% Variance explained by species
SpeciesResidual
  1. gs, stomatal conductance; gm, mesophyll conductance; Vc,max, maximum velocity of carboxylation; Amax, light-saturated maximum photosynthetic rate at ambient CO2 concentration; LMA, leaf mass per area; T, leaf thickness; D, bulk leaf density; Sc, area of chloroplasts exposed to internal air space per leaf area; SR, stomatal ratio.
Physiological   
log(gs)6.03 × 10−36.90 × 10−28.03%
log(gm)00.2620%
 Vc,max08120%
 Amax018.20%
Anatomical   
LMA31.826.354.8%
log(T)6.46 × 10−23.47 × 10−265.1%
log(D)8.32 × 10−36.15 × 10−211.9%
log(Sc)3.83 × 10−23.93 × 10−249.4%
SR4.48 × 10−23.09 × 10−259.2%

Gas exchange

We used an open-path infrared gas exchange analyser with a 2 cm2 leaf chamber fluorometer (LI-6400-40, Li-Cor Inc., Lincoln, NE, USA) to simultaneously measure leaf gas exchange and chlorophyll a fluorescence. To minimize leaf position and age effects, all measurements were made on young, fully expanded leaves (math formula, range = [4, 7]). The ambient CO2 concentration in the chamber (Ca) was 370 μmol CO2 mol−1 air, leaf temperature was 25 °C, photosynthetic photon flux density (PPFD) was 1000 μmol m−2 s−1 with 90:10 red:blue light, and relative humidity was between 50 and 70%. Once a leaf reached steady-state photosynthesis (A) and stomatal conductance (gs), usually after ∼30 min, we measured the photosynthetic light response to obtain an estimate of a leaf's maximum photosynthetic rate (Amax). The photosynthetic light response was measured between PPFD of 1500 and 25 μmol m−2 s−1 (90:10 red:blue). Using non-linear least squares regression, we fit the data to a non-rectangular hyperbola:

display math(1)

The response to changing substomatal CO2 concentrations (Ci) was measured by adjusting the ambient CO2 concentration in the leaf chamber (CA) to concentrations ranging between 50 and 1300 μmol CO2 mol−1 air. The flow rate was 300 μmol s−1. Diffusional leaks for both CO2 and H2O were corrected by using the methods of Rodeghiero et al. (2007). We observed no differences between diffusion coefficients calculated for leaves of different species, dried leaves, lyophilized leaves, or an empty chamber (data not shown).

From fluorescence measurements and A – Ci curves, we determined the photosynthetic rate, stomatal conductance (gs), mesophyll conductance to CO2 (gm) and the maximum rate of carboxylation (Vc,max). Photosynthetic rate and stomatal conductance are estimated directly from gas exchange measurements. We estimated mesophyll conductance using the variable J method (Harley et al. 1992):

display math(2)

For each leaf, gs and gm were not constant, but decreased at higher Ci, as seen in many studies (Tazoe et al. 2011 and references therein). We report gs and gm at ambient Ca (370 ± 5 μmol CO2 mol−1 air) because these values are most ecologically relevant and allowed us to directly analyse how gs and gm limited Amax, which was also measured at ambient Ca (see above). gm is sometimes averaged over a larger range of Ci, but we found that average gm was highly correlated with point measurements of gm at ambient Ca, and hence the choice of method did not qualitatively change the relationship between gm and leaf anatomy. For high gm leaves such as tomatoes, gas exchange measurements can be sensitive to error. Points measurements of ambient gm met Harley et al.'s (1992) criterion (10 > dCc/dA > 50) in all cases, indicating that these measurements were reliable.

We estimated non-photorespiratory CO2 evolution in the light (Rd) and the chloroplastic CO2 compensation point (Γ*) using the Laisk method (Laisk & Oja 1998). Briefly, for each plant, we measured the photosynthetic response to Ci over the linear portion of the response curve (generally, 30 < Ci < 120 μmol CO2 mol−1 air) at two irradiances (75 and 500 μmol m−2 s−1). We fit linear regressions to the responses for each irradiance and calculated their intersection to obtain estimates of Rd and Ci*, the intercellular CO2 compensation point, for individual leaves. Since

display math(3)

the y-intercept of the line fit between Rd and Ci* is an estimate of Γ* (Evans & von Caemmerer 1991). As there is not a causal relationship between Rd and Ci*, we calculated Γ* using type II regression, which estimates the regression slope and intercept by minimizing error in both x- and y-axes, using the R package smatr version 3.2.6 (Warton et al. 2012).

The electron transport rate of photosystem II (Jf) was calculated from fluorescence measurements as Jf = ΦPSIIIαβ, where ΦPSII is the quantum yield (moles of CO2 fixed per mole of quanta absorbed) of photosystem II, I is irradiance, α is the leaf light absorptance and β is the photosystem partitioning factor. ΦPSII was estimated from the chlorophyll fluorescence data. We estimated the product αβ using the relationship math formula under non-photorespiratory conditions (Genty et al. 1989). To vary the quantum yield, we measured photosynthetic light response curves under 2% O2. The relationship became non-linear at low irradiance as stomatal conductance decreased. Since non-linearities can bias estimates αβ and therefore gm (Gilbert et al. 2012), we fit the curve using linear regression with only measurements over the range 300 < PPFD < 1500 μmol m−2 s−1. Error in our estimates of Γ* and/or αβ could have potentially large effects on our inference about gm. To examine the sensitivity of our results to measurement error, we reran our analyses using the 0.025 and 0.975 quantiles of Γ* and αβ.

Using the calculated values of gm, we determined A – Cc curves to estimate the maximum velocity of carboxylation [Vc,max (μmol m−2 s−1)], which is proportional to the concentration of activated ribulose 1·5-bisphosphate carboxylase/oxygenase (Rubisco). We estimated Vc,max following Long & Bernacchi (2003) from the linear, Rubisco limited portion of the A – Cc curve, empirically determined to be Ci < 200 mol CO2 mol−1 air. We used constants from Sharkey et al. (2007) to calculate an effective Michaelis–Menten constant of Km = 81.15 Pa.

Leaf morphology and stomatal density

Morphological measurements were made on the same leaflets as gas exchange measurements. To calculate LMA, we used a 1 cm2 cork borer to remove a constant area of laminar tissue from every leaf. The leaf samples were dried at 60 °C and dry mass determined with a Sartorius CP225D fine balance (Sartorius Corp., Edgewood, NY, USA). LMA was calculated as g dry mass divided by the area of fresh tissue. Since LMA is the product of thickness and density, we estimated bulk leaf density (D) by dividing LMA by leaf thickness (T) measured from leaf cross sections (estimated as we explain below). A limitation of this method is that any measurement error in LMA or T will be attributed to variation in D. This causes some spurious autocorrelation between D and T. However, for this to be a serious problem, the amount of measurement error would have to be comparable to the true biological variation in T and LMA, which is unlikely given the precision with which T, dry mass and leaf area were measured. Leaf surface impressions were made by applying a thin layer of nail polish. The nail polish was removed using transparent adhesive tape and mounted on a microscope slide. Stomata were counted from three fields each of the abaxial (‘lower’) and adaxial (‘upper’) leaf surfaces. The SR was calculated as ratio of the stomatal densities (SD) on the adaxial (‘upper’) and abaxial (‘lower’) side:

display math(4)

Leaf anatomy

Following gas exchange measurements, leaf pieces from within the gas exchange chamber were fixed (2% glutaraldehyde, 1% formaldehyde; 0.05 M PIPES buffer), embedded in firm Spurr's resin, and cross sections (100–160 nm) were made using a microtome (Sorvall, Porter-Blum MT2, Norwalk, CT, USA). Images of cross sections were captured with a Nikon E800 microscope (Nikon USA, Melville, NY, USA) using bright field illumination. We calculated leaf thickness and Sc following Evans et al. (1994). Briefly, we used camera lucida drawings to measure the area and width of the sections. Leaf laminar thickness (T) was calculated as leaf cross-sectional area divided by the width of the section.

The number of chloroplasts facing intercellular air space in both palisade (Np) and spongy mesophyll cells (Ns) was separately determined for each section. The length of chloroplast touching the plasma membrane appressed to airspace was measured on five representative chloroplasts in both the palisade (Lp) and spongy mesophyll (Lsp) cells. Length measurements were made in ImageJ version 1.45 s (Schneider et al. 2012) using the segmented line tool. Sc was then calculated as

display math(5)

where Lsec is the length of the section examined. Finally, we measured fIAS, the fraction of mesophyll occupied by IAS in micrographs.

Statistical analysis

Because of shared evolutionary history, species are not necessarily statistically independent units (Felsenstein 1985). To determine whether we should incorporate phylogenetic distance into our statistical analyses, we compared the fit of a Brownian motion (BM) and white noise (WN) model of trait evolution, for each measured trait. A better fit of the BM compared with the WN model indicates that phylogenetic relatedness is positively associated with trait values and should be incorporated into the analysis. We estimated model parameters using maximum likelihood methods implemented in the R package geiger version 1.3-1 (Harmon et al. 2008) on a phylogenetic tree from Rodriguez et al. (2009). We compared model fits using a parametric bootstrapping method (Boettiger et al. 2012) implemented in the R package pmc version 0.0–8 (Boettiger 2013). For all traits, the WN model fit the data as well or better than the BM model, indicating that phylogenetic relationships did not contribute significantly to trait variance (i.e. there was no significant phylogenetic signal in the measured traits). Therefore, species were treated as independent units in the statistical analysis.

Replicates within species likewise might not be statistically independent. Since the number of replicates per species was low (math formula) and unbalanced, we used linear mixed models (LMM) treating species as a random effect. Physiological (gs, gm, Vc,max, Amax) and anatomical (LMA, T, D, Sc, fIAS, SR) traits were treated as fixed effects when used as predictor variables. For all analyses, we log-transformed response variables as necessary to normalize residuals, linearize relationships and/or decrease the leverage of a few data points. Elsewhere in this study, we generally do not log-transform the same variables when they are predictor variables and there is no assumption of normality unless it was desirable for linearization or leverage. Following the recommendations of Bolker et al. (2009), we fit LMMs using restricted maximum likelihood (REML) implemented in the lmer function, part of R package lme4 version 0.999375-42 (Bates et al. 2011). We used two approaches to assess statistical significance of fixed effects. Firstly, we used F-tests, calculating the residual degrees of freedom using the Kenward–Roger approximation (Kenward & Roger 1997) as implemented in the R package pbrktest version 0.3-1 (Halekoh & Højsgaard 2012). A significant F-test indicates that a statistical model that includes the term of interest explains more variance than expected by chance. In addition to F-tests, we determined 95% highest posterior density (HPD) intervals calculated from 104 Markov chain Monte Carlo simulations of fitted models. An HPD interval that does not overlap 0 indicates that a parameter is significantly different from 0. Note that it is possible for a term to be significant by one criterion but not the other.

We applied these statistical methods to our four questions as follows: (1) to assess the amount of interspecific variation for our physiological and anatomical traits, we compared the variance explained by species, as estimated by REML, with the total trait variation for each trait. (2) We used quantitative limitation analysis (Jones 1985, Grassi & Magnani 2005; Tomás et al. 2013) to partition controls on photosynthesis into stomatal (ls), mesophyll (lm) and biochemical (lb) limitations:

display math(6)
display math(7)
display math(8)

Following Tomás et al. (2013), we estimated ∂A/∂Cc using the slope of the A – Cc response curve between Cc = 40 and Cc = 110 μmol CO2 mol−1 air. We complemented limitation analysis using partial regression to assess whether gm per se influenced Amax independently of an indirect association between photosynthetic potential (measured as Vc,max), gm and Amax. (3) To determine the effect of variation in leaf mass per area on mesophyll conductance, we tested whether LMA was correlated, positively or negatively, with gm. (4) To determine the influence of focal anatomical traits (leaf density, leaf thickness and chloroplast surface area) on mesophyll conductance, we estimated the effect of D, T and Sc on gm simultaneously in a single LMM. We also allowed a quadratic term for T to test our a priori prediction that thickness might increase gm in thin leaves but decrease it as leaves become very thick. To test our prediction that SR mitigates the effect of leaf thickness on airspace resistance, we tested whether there was a positive correlation between SR and T.

Results

Determination of biochemical parameters

Biochemical parameters were used to estimate gm via the variable J method (see Eqn (2)). We estimated the chloroplastic CO2 compensation point (Γ*) to be 42.7 (95% CI: 40.1–45.3) μmol CO2 mol−1 air by regressing respiration (Rd) against the internal CO2 compensation point (Ci*) for all individual replicates (plants; Fig. 1). Note that while it is possible to estimate an average mesophyll conductance (gm) from the slope of the regression (see Eqn (3)), we did not use it for this purpose since we were interested in differences between plants rather than the average across plants. This method assumes that Γ* is constant across species, an assumption that is likely justified since Γ* is a property of Rubisco, which varies little across C3 species. Furthermore, sequences of the gene encoding the large subunit of Rubisco (rbcL) indicate minimal functional divergence across all wild tomatoes used in this study (J. Galmés, personal communication). In addition to Γ*, we calculated product of leaf absorptance and the photosystem partitioning factor (αβ). Using data from three species (Solanum arcanum, S. chilense and S. habrochaites), we estimated αβ = 0.472 (95% CI: 0.452–0.495), with little variance between species suggesting that this value was nearly constant in our experiment. Therefore, we used a single value for every plant. Assuming β = 0.5, α = 0.94, which was slightly higher than the measured absorptance 0.88 ± 0.03 in these species (data not shown).

Figure 1.

Estimating biochemical parameters for curve fitting. We calculated the chloroplastic CO2 compensation point (Γ*) as the intercept of the regression between mitochondrial respiration (Rd) and the intercellular CO2 compensation point (Ci*). (a) We fit the equation (top right corner) using type II regression. Each point represents an individual plant in our experiment for which we calculated Rd and Ci* using the Laisk method (see Materials and Methods for details). The larger open circle is the data point for the representative example calculation illustrated in b. Rd and Ci* were measured individually for each plant in the experiment as the intersection of CO2 response curves conducted at two light intensities [photosynthetic photon flux density (PPFD) = 75 and 500 μmol m−2 s−1].

Intraspecific and interspecific trait variation

Our first objective was to determine whether there was significant variation in physiological and anatomical traits, and how trait variation was partitioned within and among species. Physiological parameters were measured using A – Ci and light response curves (Fig. 2). Although there was substantial variation in physiological traits, most of the variation was within rather than between species (i.e. among plants; Table 1). For three of four traits, including gm, the REML estimate of species' contribution to trait variance was 0. In comparison, we detected substantial interspecific variation in anatomical traits (Table 1). Interestingly, several of these traits were correlated with gm (see below); we consider the meaning of this apparent contradiction in the Discussion section. Standard analyses of variance (anovas) treating species as a fixed rather than as a random effect gave the same qualitative results as LMMs (data not shown).

Figure 2.

Representative ACi (●; upper x-axis) and photosynthetic light response (○; lower x-axis) curves for a single plant. The ACi curve captures the response of photosynthesis (A; y-axis) to intercellular CO2 concentration (Ci; upper x-axis). The line was fit using the LOWESS algorithm (Cleveland 1979) and is intended solely for illustration rather than analysis. The light response curve captures the response of photosynthesis (A; y-axis) to photosynthetic photon flux density (PPFD; lower x-axis). The line was fit to a non-rectangular hyperbola (Eqn (1)) using non-linear least squares.

How much does mesophyll conductance limit photosynthetic rate?

In general, biochemistry limited photosynthetic rate more than gm, and gm limited photosynthetic rate more than gs (Table 2). Across plants, biochemical (lb) were on average greater than stomatal (ls) limitations (paired t-test, t42 = 11.97, P = 4.00 × 10−15) and mesophyll (lm) limitations (paired t-test, t42 = 3.58, P = 0.0009). Likewise, lm was on average greater than ls (paired t-test, t42 = 5.16, P = 6.42 × 10−6). However, the relative limitation imposed by mesophyll versus stomatal conductance was sensitive to αβ (Supporting Information Fig. S3), indicating that this result might not be robust to measurement error. We also used partial regression to examine whether variation in gm per se affected photosynthetic performance, which we measured as the maximum photosynthetic rate (Amax) under saturating irradiance (Eqn (1)). An alternative hypothesis is that gm is correlated with photosynthetic potential, leading to an ostensibly strong but indirect relationship between gm and Amax. For fitting purposes, gm was log-transformed prior to analysis to produce a linear relationship with Amax. gm and Vc,max, an indicator of photosynthetic capacity, were both associated with greater Amax (Fig. 3a,b), but were uncorrelated with each other (Fig. 3c). The partial correlation between gm and Amax, conditioned on Vc,max (ρ = 0.84, t41 = 10.08, P = 1.17 × 10−12), confirms that the relationship between gm and Amax is not mediated by Vc,max.

Figure 3.

Pairs plot of physiological traits. Each data point is an individual plant from the species indicated by the legend. Maximum photosynthetic rate (Amax) is correlated with (a) mesophyll conductance (gm) and (b) the maximum rate of carboxylation (Vc,max). (c) gm is not correlated with Vc,max. The linear mixed model (LMM) regression lines were fit by restricted maximum likelihood (REML), treating species as a random effect.

Table 2. Quantitative limitation analysis comparing stomatal (ls), mesophyll (lm) and biochemical (lb) limitations to photosynthetic rate. Note that, by definition, ls + lm + lb = 1. In wild tomatoes, ls < lm < lb in non-stressed conditions
LimitationStomatal (ls)Mesophyll (lm)Biochemical (lb)
Mean ± SEM (n = 44)0.24 ± 0.0070.33 ± 0.0150.43 ± 0.014

How does LMA affect mesophyll conductance?

LMA was negatively associated with mesophyll conductance (Fig. 4a, F1,27.81 = 7.16, P = 0.012), indicating that even in herbaceous tomato leaves, increased investment in leaf tissue (higher LMA) significantly inhibits gm. The 95% HPD interval for the coefficient relating LMA to gm does not overlap 0 (−1.41 – −0.23). For these results, we log-transformed both gm and LMA prior to analysis because both traits were log-normally distributed, meaning that residuals are highly heteroscedastic. For comparison, the relationship is significant even if LMA is not log-transformed (F1,25.38 = 8.28, P = 0.008) and marginally significant if neither variable is transformed (F1,19.56 = 3.98, P = 0.060). The CO2 drawdown from substomatal cavities to the chloroplast (CiCc) was also significantly positively correlated with LMA (Fig. 4b, F1,28.60 = 9.15, P = 0.005). The 95% HPD interval for the coefficient relating LMA to CO2 drawdown does not overlap 0 (19.41–89.91). The effect of LMA of gm and CO2 drawdown was not sensitive to measurement error in Γ* or αβ (Supporting Information Fig. S4). This provides further evidence that greater LMA hinders internal CO2 diffusion. We also reanalysed a previously published global dataset of LMA, gm and CO2 drawdown (Niinemets et al. 2009a) to put our results in context. The slope of the log(gm)–log(LMA) relationship in the global dataset (−0.66, 95% CI: −0.96 – −0.37) fell within 95% HPD interval determined from tomatoes (see above), indicating a general pattern of declining in gm with increasing LMA (Supporting Information Fig. S1a). The slope of the global (CiCc)–log(LMA) likewise fell within the interval determined from tomatoes (26.31, 95% CI: 11.97–40.69), although there was less overlap between datasets in this comparison than that with gm (Supporting Information Fig. S1b).

Figure 4.

Leaf dry mass per area (LMA) decreases mesophyll conductance (a) and CO2 drawdown from substomatal cavities to chloroplasts (b). (a) Mesophyll conductance (gm; y-axis) is negatively correlated with LMA (x-axis). (b) Decreased gm in higher LMA leaves lead to a greater CO2 drawdown (CiCc; y-axis). Each data point is an individual plant. Following Fig. 3, the different symbols indicate a different species. The linear mixed model (LMM) regression line was fit by restricted maximum likelihood (REML), treating species as a random effect.

What is the mechanistic basis of the LMA–gm relationship?

LMA is equal to the product of leaf thickness (T) and density (D). Therefore, the relationship between LMA and gm must be mediated by these traits. We hypothesized that leaf thickness could be associated with increased gm in herbaceous leaves if thicker leaves have increased chloroplast surface area (Sc), a trait often positively correlated with gm. Although T and Sc were positively correlated with each other (Supporting Information Fig. S2b, F1, 29.09 = 9.24, P = 0.005), neither was significantly positively correlated with gm (Table 3, but see below). In fact, we detected no effect of Sc but a significant negative effect of T according to both F-tests and 95% HPD intervals (Table 3). We log-transformed T before analysis to reduce the leverage of a few S. pennellii individuals with exceptionally thick leaves and low gm. Using untransformed T makes the result more significant (F1, 14.68 = 6.16, P = 0.026). Although the primary effect of T on gm was negative, there was some evidence of a positive association between T and gm at very low T (approx. T < 125 μm). Specifically, a statistical model including quadratic and linear terms for T (Supporting Information Table S3) fit better than a model including only the linear term (Supporting Information Table S2) according to Akaike information criteria (AIC). However, the HPD intervals for the quadratic term overlapped 0 (Supporting Information Table S3) and this term was not significant according to F-tests (F1, 18.23 = 1.49, P = 0.237). Leaf density (D) had a stronger effect on gm than T (Table 3). The effect of D might be mediated by changes in leaf porosity, as D was negatively correlated with the airspace fraction, fIAS (F1, 25.49 = 10.20, P = 0.004). However, fIAS itself was not significantly correlated with gm (F1, 29.24 = 1.24, P = 0.27), indicating the changes in the liquid phase resistance were probably also relevant. Finally, the fact that leaf thickness did not affect mesophyll conductance as strongly as leaf density might be in part because SR was positively correlated with thickness (Fig. 5), reducing the airspace resistance in the thickest leaves by increasing the proportion of stomata on the upper surface of the leaf.

Figure 5.

Thicker leaves have more stomata on the adaxial (upper) surface. Stomatal ratio (SR; y-axis) is positively correlated with leaf thickness (T; x-axis). SR is defined as the ratio of the adaxial to the abaxial stomatal density (see Eqn (5)). Each data point is an individual plant. Following Fig. 3, the different symbols indicate a different species. The linear mixed model (LMM) regression line was fit by restricted maximum likelihood (REML), treating species as a random effect.

Table 3. Linear mixed model (LMM) table of anatomical traits and mesophyll conductance. Bolded terms are statistically significant fixed effects according to two criteria: they increase model fit (significant F-test) and are significantly different from 0 [95% highest posterior density (HPD) interval does not overlap 0]
log(gm) ∼ REML estimate95% HPD interval  
  1. T, leaf thickness; D, bulk leaf density; Sc, area of chloroplasts exposed to internal air space per leaf area.
Random effects    
Species0.0210–0.450  
Residual0.1830.108–0.325  
Fixed effects  Fdf1, df2P
(intercept)3.72−0.879–7.98
 D−5.67−9.57 – −1.56F1, 27.97 = 7.140.012
 log(T)−0.972−1.84 – −0.125F1, 19.72 = 4.580.045
 Sc0.089−0.050–0.213F1, 24.87 = 1.480.235

Discussion

Efficient fixation of CO2 is essential for plant growth and fitness. The physics of CO2 diffusion is therefore a significant factor shaping the evolution of leaf morphology and anatomy. When comparisons are made between very broad functional groups, mesophyll conductance (gm) to CO2 appears to be an ecologically important limitation on photosynthetic performance that is determined largely by leaf structure. Greater investment in leaf tissue per area (LMA) limits photosynthetic performance by decreasing gm, especially in succulent and sclerophyllous leaves (Maxwell, von Caemmerer & Evans, 1997; Hassiotou et al. 2010; Tosens et al. 2012b). It is less clear how leaf structure affects CO2 diffusion in herbaceous leaves with relatively high gm. Here, one of our aims was to determine whether morphological and anatomical variation within and among closely related herbaceous species (wild tomatoes) affected photosynthetic performance by influencing gm.

Under the non-stressed conditions of our experiment, we found no significant differentiation between species in physiological traits (Amax, Vc,max, gs, gm) even though there were significant interspecific differences in anatomical traits (Table 1) that impact gm, especially LMA (Fig. 4). The discrepancy between physiological and anatomical traits with respect to interspecific differences probably reflects statistical limitations rather than biology. In contrast to easily measured traits like LMA, physiological traits like stomatal conductance can change rapidly in response to immediate environmental variation, leading to high trait variation among replicate plants. Given the additional variation in physiology caused by environmental fluctuations, the failure to detect strong physiological differentiation between species might reflect limited replication (math formula) and taxon sampling (eight species). Indeed, a broader analysis of physiological variation across 19 tomato species and close relatives found significant interspecific variation (Muir, unpublished data). In contrast, in this study, within-species variation in gm and other physiological traits was more important than variation between species. Therefore, our conclusions are based primarily on within species (among plant) variation resulting from physiological plasticity and/or intraspecific genetic variation.

We found that variation in gm strongly affected photosynthetic performance in wild tomatoes (Fig. 3a, Table 2). Quantitative limitation analysis showed that mesophyll limitations were generally greater than stomatal limitations, but less than biochemical limitations (Table 2). However, this conclusion is sensitive to measurement error in model parameters (Supporting Information Fig. S3). In many plants, high mesophyll conductance is coordinated with high photosynthetic capacity, suggesting that these traits might not be independent (Buckley & Warren 2013). However, we found that Vc,max and gm were uncorrelated (Fig. 3c) and partial correlation indicates that mesophyll limitations are independent of biochemical limitations.

We were also interested in examining the relationship between LMA, a highly variable leaf trait, and gm. We found that LMA and gm were negatively correlated (Fig. 4a). This negative association is consistent with other studies that have made comparisons across broad functional groups with a large range of LMA; in contrast, among species with low LMA leaves, gm has been found to be highly variable (Flexas et al. 2008). Our study, however, found that a negative relationship held even among very closely related individuals and species within this group. Combined analysis of our dataset and a previously published, taxonomically broad dataset (Niinemets et al. 2009a) suggests that a single slope adequately describes the decline in gm from low to high LMA ends of the leaf phenotypic spectrum (Supporting Information Fig. S1a). Furthermore, the CO2 drawdown from substomatal cavities (Ci) to chloroplasts (Cc) increased with LMA (Fig. 4b), indicating that traits associated with greater LMA reduce diffusion through the mesophyll independent of how stomatal conductance affects Ci. Despite the fact that gm is generally high in wild tomatoes, as in other herbaceous taxa (Flexas et al. 2008), it still strongly limited photosynthetic performance independent of both photosynthetic capacity and stomatal conductance.

Nonetheless, the negative relationship between LMA and gm appears to be mediated by a relatively complex interaction between leaf tissue density, thickness, porosity, stomatal distribution and possibly other unmeasured traits. Unexpectedly, much of the association between LMA and gm was apparently mediated by airspace rather than liquid phase resistance. Thicker leaves generally had lower gm despite the fact that thicker leaves had higher Sc (Supporting Information Fig. S2) and higher SR (Fig. 5). The decline in gm at higher T differs from the pattern seen across genotypes of cultivated tomato (Galmés et al. 2013), either because wild tomatoes vary in different ways compared to cultivars or because plants were raised under different experimental conditions. Airspace resistance might have also contributed to the negative effect of leaf density on gm. Leaf density was partially explained by increased cell packing and therefore decreased leaf porosity (fIAS). However, since fIAS per se was not a significant predictor of gm, other leaf properties, such as mesophyll cell wall thickness, must account for the negative effect of density on gm. Direct estimates of parameters associated with liquid phase resistance are needed to partition the importance of air and liquid phase resistance (Tosens et al. 2012a,b; Tomás et al. 2013). We are currently measuring these parameters in a larger survey of photosynthetic and morphological variation in this group (Muir, unpublished data). Regardless, our conclusions are consistent with biophysical models indicating that airspace resistance can set an upper limit on photosynthesis in thick leaves (Parkhurst 1994; Terashima et al. 2006; Flexas et al. 2008, Niinemets et al. 2009aa). However, our data also indicate that gm is usually determined not by a single trait, but rather by complex trait covariation (Tosens et al. 2012b, Giuliani et al. 2013).

Two results from this study appear to be inconsistent with other theory and empirical data: Sc was not correlated with gm and two lines of evidence suggested that airspace resistance was appreciable. In contrast, previous studies indicated that Sc is positively correlated with gm and that liquid phase resistance is more important (e.g. Tosens et al. 2012b), especially in thin, amphistomatous leaves. Our results could be interpreted to mean that Sc and liquid phase resistance are not important in these species. For example, Sc does not strongly affect gm when mesophyll cell walls are thick (Evans et al. 2009) and some studies similar to ours that compare gm among closely related genotypes or species fail to show a significant affect of Sc under well-watered conditions (Galmés et al. 2013; Giuliani et al. 2013). Alternatively, it is possible that Sc and liquid phase resistance limit photosynthetic performance but, because they did not vary much among our experimental individuals, they do not contribute to explaining the Amax variation observed among our study individuals. That is, although these traits might be physiologically important because they are invariant (possibly because of selection to maximize Sc and/or minimize liquid phase resistance in all species), they cannot contribute to evolutionarily significant differences among wild tomatoes.

Although further work will be needed to understand the mechanistic basis for the negative effect of LMA on gm in wild tomatoes, our data indicate that increased LMA inhibits CO2 diffusion and photosynthetic performance in these species. Increased LMA is also known to reduce intrinsic biomass growth rate (Poorter et al. 2009), indicating that increased LMA has significant costs in terms of both photosynthesis and whole plant growth. How does this cost of LMA impact leaf phenotypic evolution across climatic gradients? One hypothesis is that increased LMA confers stress tolerance and that reduced growth and photosynthetic efficiency are the unavoidable, intrinsic costs associated with this stress tolerance strategy (Niinemets et al. 2009b). However, desert annuals, drought deciduous species and herbaceous perennials can respond dynamically to ephemerally optimal conditions, and actually have remarkably high photosynthetic performance under their optimal conditions (Gibson 1998), indicating adaptation to a stressful environment is not intrinsicially associated with slow growth rates. Indeed, in plants that employ a drought escape strategy (rapid growth during high water availability, followed by dieback during drought), tender leaves with efficient CO2 diffusion might be beneficial, allowing them to take advantage of intermittent periods of water availability. For example, in Populus balsamifera, it has been hypothesized that high-latitude genotypes with shorter growing seasons have been selected to have greater photosynthetic performance via increased gm specifically during their limited optimal growing season (Soolanayakanahally et al. 2009). This suggests that physiological plasticity in response to changing environmental conditions might be a critical factor in determining the nature of the relationship between LMA, photosynthetic performance and adaptation to resource limited environments, and therefore, the relationship between leaf phenotypes and climatic variation. The focus of this study was on functional relationships, and our limited taxon sampling did not permit us to investigate these ecological hypotheses here. A broader study of physiological and anatomical variation in this group, currently under way, will enable us to address these questions and better understand the mechanistic basis of adaptation to different environments.

Acknowledgments

We would like to thank Barry Stein at the Indiana University Light Microscopy Imaging Center for assistance with microtomy. Two anonymous reviewers provided thoughtful comments that improved the quality of this manuscript. C.D.M. was supported by an NSF Graduate Research Fellowship. We have no conflicts of interest to declare.

Ancillary