Angiosperm leaf vein patterns are linked to leaf functions in a global-scale data set†
The author thanks the Peabody Museum of Natural History at Yale University for access to cleared leaf specimens, L. Hickey and S. Hu for assistance in accessing the collection, D. Briggs for the use of his microscope and laboratory, I. Wright and P. Reich for granting permission to use the Glopnet database, G. Bolen for assistance with data collection, and R. Geeta, J. Wiens, M. Lerdau, L. Sack, and three anonymous reviewers for comments on earlier versions of this manuscript.
Abstract
• Premise of the study: Leaves are plants’ primary interface with the atmosphere and affect a range of ecological processes. Vein patterns are one of the most prominent aspects of leaf form, and the functional significance of different vein patterns is gaining increasing attention.
• Methods: Phylogenetic and standard ANOVA and regression were used to provide the first global-scale, phylogenetically based test of relations between angiosperm vein patterns and leaf functional traits. Pagel's λ was used to test for phylogenetic signal in all traits.
• Key results: All leaf traits had significant phylogenetic signal. Significant phylogenetically based relations were found between secondary vein pattern and leaf functions, linking leaf form to the well-known trade-off between physiological activity and leaf life span. The relations between primary vein pattern and leaf functions were not found to be significant with phylogenetic tests, suggesting these relations may be the result of changes within a few lineages, followed by phylogenetic conservatism, rather than multiple instances of correlated trait evolution. The relation between minor vein density and maximum photosynthetic rate was found to be marginally nonsignificant with phylogenetic regression, which does not rule out coordinated evolution of hydraulic supply and demand.
• Conclusions: Although phylogenetic conservatism may weaken statistical relations between vein patterns and leaf functions, phylogenetic relations can provide a complementary source of information for inferring unmeasured values of leaf traits. Relations among vein patterns, leaf functions, and phylogeny will be valuable for estimating functional attributes of living and fossil plant species and communities.
As the primary interface between plants and the atmosphere, leaves affect a range of ecological processes, from herbivore dynamics to ecosystem productivity. The study of leaf traits has become an active subdiscipline within ecology, providing insights into how leaf forms affect ecological and evolutionary processes (e.g., 16; 37; 36; 72). Veins are one of the most visible traits of leaves; they provide support, water delivery, and carbohydrate export and are crucial for maintaining leaf water status and photosynthetic capacity (55; 61). The past decade has seen many new studies on the relations between vein characteristics and leaf functions (e.g., 63; 62; 12; 45; 59; 10; 44), yet there is much more to learn, particularly about the functional differences among major vein patterns. Although phylogenetic information is increasingly being incorporated into studies of leaf traits within genera (e.g., 22; 21), phylogenetic comparative methods are rare in broad-scale leaf trait studies (e.g., 2; 11). This makes it difficult to determine whether the relations between leaf veins and leaf functions that have been identified to date arose through multiple instances of correlated evolution or are the result of phylogenetically conserved trait relations within large clades (25; 33).
In this study, I used phylogenetic comparative methods to examine relations between angiosperm vein patterns and the leaf economic spectrum (LES; 77) with the use of new data on vein patterns and the global-scale data set of leaf traits from 77. The LES illustrates a trade-off between long leaf life span (LLS) and high physiological activity by demonstrating that across thousands of species from a range of biomes and growth forms, relations among LLS, maximum photosynthetic rate (Amax), leaf nitrogen (N) concentration, and leaf mass per area (LMA) fall along one multivariate axis (66; 52; 53; 77). The traits described by the LES are known as leaf economic traits (52) because they are based on an optimization principle that partitions investment of scarce resources into either support/strengthening tissue or physiologically active tissue. Leaf economic traits are widely studied because of their importance for both plant functions and ecosystem processes (e.g., 37; 43; 65; 72; 1; 30; 39; 57).
Several relations between vein patterns and leaf economic traits already have been found at smaller spatial or phylogenetic scales or with smaller data sets. The goals of this study were to determine whether those relations are present in a global-scale data set and to test whether they arose from repeated, correlated trait evolution. I included three different vein patterns: primary vein type, secondary vein type, and minor vein density (MVD). Primary and secondary veins are considered major veins and act as the support and distribution network for leaves (38; 55). Primary veins are classified as pinnate, palmate, or parallel (Fig. 1A). Secondary veins branch from the primaries and can be divided into closed forms such as brochidodromous veins, open forms such as craspedodromous veins, and intermediate forms such as eucamptodromous veins (Fig. 1B). The 38 classified third-order and higher veins as minor veins, though other researchers have found that only fourth-order and higher veins are minor veins (e.g., 60), and still others have not clearly defined what they mean by minor veins. Minor veins act as the sites of exchange between the mesophyll and the vascular system (31; 61). They have variable arrangements (38), but little is known about the functional significance of these arrangements (but see 7). I focused on MVD (the length of minor veins per area), since the functional significance of this trait has been examined by a number of researchers but rarely with phylogenetic methods (69; 60; 12; 47; 10; 44).
Schematic representation of primary and secondary vein patterns. (A) Pinnate-veined leaves have single primary vein running the length of the leaf, palmate-veined leaves have multiple primary veins that radiate from the petiole, and parallel-veined leaves have multiple primary veins that run roughly parallel from the petiole to the leaf apex. (B) Closed secondary veins connect at the leaf margin. Open secondary veins end at or near the leaf margin and do not connect directly to other secondary veins. Intermediate secondary veins curve at their apex to run parallel to the leaf margin but do not directly connect to other secondary veins. These illustrations represent possible leaf shape/vein type combinations, but multiple shapes exist within each category.
Largely on the basis of the results of two previous, nonphylogenetic studies, I predicted that species with palmate or parallel venation would be associated with the high physiological activity end of the LES and species with pinnate venation with the low end. In the first study, 45 showed that among 44 species from two temperate forest sites, pinnate leaves had lower primary vein density than palmate or parallel-veined leaves, but they invested more biomass in support tissue outside the primary vein, with higher MVD and carbon (C) concentration. These differences suggest that pinnate-veined leaves should have higher LMA but lower N concentration and Amax. The second study (59) compared seven temperate species and found that palmate-veined leaves maintained higher leaf hydraulic fluxes than pinnate-veined leaves after their midribs were severed, due to the vascular redundancy conferred by higher primary vein density. Parallel-veined leaves were not tested in that study, but I hypothesized that they would have redundancy similar to that seen with palmate venation because of their high primary vein density. 59 hypothesized that the damage tolerance conferred by the vascular redundancy of palmate venation would be most advantageous for thin leaves that lack the alternative form of protection offered by thick, tough laminas. On the basis of these earlier studies and the relations established by the LES (66; 52; 53; 77), I predicted that leaves with palmate or parallel venation would have lower LMA, higher Amax, and higher N concentration than would leaves with pinnate venation.
Few empirical studies exist that examine the functions of secondary veins, so the predictions for secondary veins were based largely on theoretical work (29; 35; 54; 46; 55; 8). Closed secondary vein patterns reinforce the leaf edge, reducing the likelihood of tearing (46; 55), and may provide a more even distribution of water for improved physiological functioning near the lamina margin (54), particularly in the event of damage or under conditions of fluctuating flow (17; 34). Open vein patterns theoretically can supply a given leaf area using the least amount of vascular tissue but with less even water distribution (35; 54; 8). Because of the putative costs associated with extra vein tissue at the leaf margin, I predicted that closed secondary veins would be associated with species that have long-lived leaves (with higher LMA and lower N) that have more time to recoup the cost of building marginal veins. I predicted that leaves with open secondary veins (and putatively less secondary vein tissue) would be associated with species that have short LLS and would use a strategy of higher investment in physiologically active tissue for fast returns (low LMA and high N). This prediction is also consistent with the global distribution of leaf margin types. Entire-margined leaves are predominant in wet tropical climates with many evergreen species, whereas toothed or lobed leaves are common in temperate climates with many deciduous species (5; 76; 73; 56), and my preliminary examinations suggested an association between open vein patterns and toothed or lobed margins and between closed or intermediate vein patterns and entire margins.
A positive relation between MVD and leaf-level fluxes has been demonstrated in several nonphylogenetic studies. Minor vein density was correlated with hydraulic conductance in tropical angiosperms (60) and with transpiration rate in land plants (10). The distance from minor veins to stomata, for which MVD is a proxy (69; 47), also was correlated with Amax and hydraulic conductance in land plants (12). These relations are thought to represent coordinated evolution of leaf hydraulic capacity and photosynthetic capacity (61; 12), but such a hypothesis requires explicit phylogenetic testing. Therefore, I used phylogenetic comparative methods to examine the relation between MVD and maximum photosynthetic rate. Even when measured on the same leaves, factors such as leaf thickness and mesophyll structure may confound the relation between MVD and fluxes. However, if the covariation between hydraulic supply and demand in contemporary species is a result of correlated evolutionary changes, then these functions should be species-level properties. Therefore, I tested whether MVD from one sample of leaves could be used to predict Amax from an independent leaf sample of the same species, despite intraspecific variation, using both phylogenetic and nonphylogenetic regression. Significant results for these tests would lend extra credibility to estimates of physiological traits in fossil taxa based on MVD (9; 10; 11).
To test the predictions laid out above, I collected data on major vein patterns for over 400 species of angiosperms and on MVD for approximately 100 species, and I used published data on leaf economic traits (77). I used the program Phylomatic (71) to generate phylogenies for the species included in this study, and I used standard and phylogenetically based ANOVA and regression analyses to test all relations. Major vein patterns are generally conserved within genera and families (68; 20), but little is known about phylogenetic patterns of minor veins (but see 12; 10), and no broad-scale tests of phylogenetic signal in leaf economic traits exist (but see 2). Therefore, I used Pagel's λ (48) to test for phylogenetic signal in all traits before I examined relations among traits.
MATERIALS AND METHODS
Data collection
I collected data on major vein patterns for all species for which Amax or LLS data were available from the Glopnet database (77) and for which images that clearly showed the major veins were available from online image collections of herbaria and botanical gardens. This included 468 species for primary vein type and 361 species for secondary vein type. I scored leaves for primary and secondary vein types according to Fig. 1. Data on MVD came from 96 species from the National Cleared Leaf Collection at the Peabody Museum of Natural History. I used a MZ16 microscope (Leica Microsystems, Bannockburn, Illinois, USA) to record digital images of leaf veins at 40× magnification. Using ImageJ version 1.41o (51), I cropped leaf images to an area of 0.25 mm2, centering images between third-order veins, and measured the length of all fourth- and higher-order veins. I calculated MVD as total vein length/area. I measured MVD for four areas per leaf, all located near the midrib, approximately midway from the base to the tip of the leaf, and used the average of those four areas. When multiple leaves were available for a species (29 of 96 species), I measured MVD for each leaf and calculated the species average. Data on leaf economic traits came from 77. For comparisons with major vein patterns, this included data on Amax and N concentration on a mass basis (Amass and Nmass), LMA, and LLS. For comparison with MVD, I used Amax on an area basis (Aarea) because MVD is an area-based measurement, and area-based fluxes have been used in previous studies (12; 47). When more than one entry per species existed in the Glopnet database, I calculated the species average value of each trait. All variables were log10 transformed before analysis.
Phylogenies
I checked species and genus names against the International Plant Names Index (http://www.ipni.org:80/ipni/plantnamesearchpage.do, accessed April 2009) or Tropicos names database (http://www.tropicos.org/NameSearch.aspx, accessed April 2009) and added family names from the same sources. I used the online software package Phylomatic (71) to construct phylogenies for each data set (primary vein type, secondary vein type, and MVD) by pruning supertrees to contain only the sampled species. Several master trees are available through Phylomatic, each representing a slightly different phylogenetic hypothesis. The Angiosperm Phylogeny Group 3 (APG3) master tree (3) is the most up to date but has no option to include branch lengths. The master tree based on the phylogeny by 18 also has no branch lengths. The conservative master tree based on the early Angiosperm Phylogeny Group study (APG1; 67) is outdated, but it provides the option of adding branch lengths by using the BladJ function in Phylomatic, based on dated nodes from 75. Despite the uncertainty, these branch lengths are more realistic than the alternative of equal branch lengths because analyses based on trees with equal branch lengths are highly sensitive to incomplete taxon sampling (41). All master trees are resolved to the family level, with additional resolution within some families provided by the authors of Phylomatic. To examine the effects of phylogenetic uncertainty, I ran my analyses using four master trees: APG3, the Davies et al. tree, APG1 with branch lengths, and APG1 without branch lengths. I used Mesquite version 2.6 (40) to randomly resolve all polytomies to zero branch lengths and to prune trees as necessary to account for missing data in some traits.
Statistical analyses
To test for phylogenetic signal, I determined how well the distribution of actual trait values for each of the four master trees fit different models that do or do not include phylogenetic covariation. Continuous traits were modeled with the “fitContinuous” command from the Geiger package version 1.2-14 (32) in R version 2.8.1 GUI version 1.27 Tiger for Macintosh (70) to determine whether the distribution of each variable fit best to a model of white noise (all trait values drawn from the same normal distribution, equivalent to no phylogenetic signal), a Brownian motion model (variables evolve along the tree following a random walk model), or the λ model, which is Brownian motion with an additional parameter λ. Lambda scales all internal branches of the phylogeny such that λ = 0 corresponds to a star phylogeny (no phylogenetic signal), whereas λ = 1 indicates that a trait is evolving according to the random walk model (48). Maximum likelihood (ML) values of λ significantly greater than 0 but less than 1 indicate that the Brownian motion model overestimates the correlation between phylogeny and trait values. A better fit to either the Brownian motion or λ model indicates phylogenetic signal. Discrete traits were analyzed with the “fitDiscrete” function in Geiger, which uses continuous-time Markov models to fit trait values to different models of evolution. I compared models with λ set to 0 (no phylogenetic signal), to 1 (phylogenetic signal, and trait evolving according to Brownian motion), and to the ML value (0 < λ < 1, interpreted as previously described).
To test for differences in Amass, Nmass, and LMA among primary vein types and for differences in LLS, LMA, and Nmass among secondary vein types, I conducted standard ANOVA on the raw data, using the R function “aov” with Tukey's honestly significant difference test, as well as phylogenetic ANOVA (26) as implemented with the “phy.anova” command in the R package Geiger (32). Phylogenetic ANOVA takes phylogenetic covariation into consideration by using simulations to assess the likelihood of arriving at a given distribution of traits by chance, assuming a Brownian motion model of evolution (λ = 1). For trait/tree combinations that fit best to a model of evolution with 0 < λ < 1 (see Results), I transformed the tree by multiplying branch lengths by the ML value of λ of the continuous trait before conducting the phylogenetic ANOVAs. I used a χ2 test to determine whether differences existed in the proportions of different major vein types in different habitats, based on species’ biome classifications from Glopnet (77).
To test hypotheses relating to minor vein patterns, I used the R function “ls” in the nlme package to conduct phylogenetic generalized least squares (PGLS) regression (42) of Aarea on MVD while simultaneously estimating the ML value of λ for the tree, considering both traits. For comparison, I also conducted PGLS regression with λ set to 1, which is equivalent to a Brownian motion model and gives the same results as phylogenetic independent contrasts (25). Regression analysis was performed on all four of the master trees. For nonphylogenetic analysis, I used GLS regression with λ set to 0, which is equivalent to ordinary least squares (OLS) regression. As an indication of the explanatory power of the regression models, I reported r2 from the OLS regression.
RESULTS
The phylogenies used in this study are available in Appendices S1–S12 (see Supplemental Data online at http://www.amjbot.org/cgi/content/full/ajb.1000154/DC1). Phylogenetic signal existed in all leaf economic traits and vein patterns, as indicated by ML values of λ that were significantly greater than 0 for all traits for all trees (Tables 1, 2). The ML value of λ was equal to or close to 1 for all traits for all trees. The most likely model did not differ among the trees without branch lengths (APG1 without branch lengths, Davies et al. tree, or APG3 tree) for any trait, suggesting that tree topology had no effect on the results for these trees (Tables 1, 2). Including branch lengths did affect the choice of evolutionary models for the primary vein type data set: using the APG1 tree with branch lengths, the ML value of λ was significantly less than 1 for all three traits, whereas it was equal to 1 using the APG1 tree without branch lengths (Table 1). The choice of evolutionary models also may be sensitive to taxon sampling, since the ML values of λ for LMA and Nmass were significantly different from one using the secondary vein data set, but not using the primary vein data set, for trees without branch lengths (Table 1). Nonetheless, all data sets consistently showed phylogenetic signal in all traits, and the inclusion of phylogenetic error structure consistently improved the likelihood of all ANOVA and PGLS models (Tables 3, 4).
| Trait (Data set) | Tree | White noise | Brownian motion | Lambda | ML value of λ |
|---|---|---|---|---|---|
| Amass (Primary vein type) | APG1 BL | –88.3547 | –84.2976 | –23.4306 | 0.8797* |
| APG1 no BL | –88.3547 | –46.0963 | –38.9171 | 0.8647* | |
| Davies et al. | –88.4345 | –49.3553 | –40.6270 | 0.8828* | |
| APG3 | –87.7002 | –42.4046 | –33.1399 | 0.8936* | |
| LMA (Primary vein type) | APG1 BL | 32.9999 | 42.1237 | 101.4663 | 0.9151* |
| APG1 no BL | 32.9999 | 80.3174 | 82.7717 | 0.9429* | |
| Davies et al. | 31.3295 | 75.0208 | 78.2710 | 0.9526* | |
| APG3 | 30.7156 | 85.2256 | 89.1811 | 0.9506* | |
| Nmass (Primary vein type) | APG1 BL | 87.5444 | 94.6030 | 182.8255 | 0.8860* |
| APG1 no BL | 87.5444 | 173.8689 | 178.6290 | 0.9315* | |
| Davies et al. | 87.1816 | 171.2984 | 176.9528 | 0.9364* | |
| APG3 | 87.5241 | 173.9758 | 180.2766 | 0.9375* | |
| LLS (Secondary vein type) | APG1 BL | –142.9710 | –120.5002 | –93.2173 | 0.9139* |
| APG1 no BL | –142.9710 | –111.5276 | –110.0860 | 0.9392 | |
| Davies et al. | 139.7575 | –111.2046 | –109.3343 | 0.9466 | |
| APG3 | –137.0778 | –97.7523 | –96.8847 | 0.9723 | |
| LMA (Secondary vein type) | APG1 BL | 3.8738 | –16.4862 | 38.9466 | 0.7638* |
| APG1 no BL | 3.8738 | 16.3869 | 16.3869 | 1 | |
| Davies et al. | 2.8127 | 9.4922 | 9.4922 | 1 | |
| APG3 | 3.5688 | 12.0675 | 12.0675 | 1 | |
| Nmass (Secondary vein type) | APG1 BL | 100.9213 | 91.8098 | 154.3407 | 0.8655* |
| APG1 no BL | 100.9213 | 133.0868 | 134.5095 | 0.9529 | |
| Davies et al. | 105.8308 | 137.1090 | 139.1399 | 0.9590 | |
| APG3 | 105.7462 | 139.2887 | 141.3851 | 0.9575 | |
| MVD (Minor veins) | APG1 BL | 53.1976 | 46.4030 | 61.5076 | 0.8269* |
| APG1 no BL | 53.1976 | 52.1247 | 54.9541 | 0.7943* | |
| Davies et al. | 55.9206 | 51.7137 | 56.7309 | 0.5745* | |
| APG3 | 55.9206 | 51.3632 | 56.6180 | 0.6039* | |
| Aarea (Minor veins) | APG1 BL | 31.4826 | 16.7481 | 44.7093 | 0.8171* |
| APG1 no BL | 31.4826 | 41.4763 | 42.1783 | 0.9373 | |
| Davies et al. | 32.8226 | 40.8066 | 41.8268 | 0.9125 | |
| APG3 | 32.8226 | 39.2411 | 39.7821 | 0.9543 |
- Note: Aarea = maximum photosynthetic rate on an area basis; Amass = maximum photosynthetic rate on a mass basis; APG1 BL = the Angiosperm Phylogeny Group (APG; 3) 1 master tree with branch lengths; APG1 no BL = APG1 master tree without branch lengths; APG3 = APG3 master tree without branch lengths; Davies et al. = 18 master tree; LLS = leaf life span; LMA = leaf mass per area; MVD = minor vein density; Nmass = nitrogen concentration on a mass basis.
| Trait | Tree | λ = 0 | λ = 1 | λ = ML value | ML value of λ |
|---|---|---|---|---|---|
| Primary vein type | APG1 BL | –293.1246 | –171.0728 | –171.0728 | 1 |
| APG1 no BL | –308.4534 | –176.4878 | –176.4878 | 1 | |
| Davies et al. | –316.5376 | –175.3941 | –175.3970 | 0.9982 | |
| APG3 | –311.3601 | –178.0245 | –178.0341 | 0.9967 | |
| Secondary vein type | APG1 BL | –396.2278 | –363.3613 | –380.4627 | 0.9161* |
| APG1 no BL | –394.1401 | –358.0072 | –358.0405 | 1 | |
| Davies et al. | –391.8256 | –356.7674 | –356.8168 | 1 | |
| APG3 | –311.3601 | –178.0245 | –178.0341 | 0.9967 |
- Note: See Table 1 for meanings of abbreviations.
| Predictor | Response | Standard ANOVA | APG1 BL | APG1 no BL | Davies et al. | APG3 |
|---|---|---|---|---|---|---|
| Primary vein type | Amass | <<0.001 (468) | 0.265 (468) | 0.272 (468) | 0.340 (462) | 0.264 (460) |
| LMA | <<0.001 (468) | 0.136 (468) | 0.157 (468) | 0.255 (462) | 0.159 (460) | |
| Nmass | <<0.001 (420) | 0.260 (420) | 0.198 (420) | 0.289 (426) | 0.206 (424) | |
| Secondary vein type | LLS | <<0.001 (361) | <0.001 (361) | <0.001 (361) | <0.001 (358) | <0.001 (354) |
| LMA | 0.002 (315) | 0.038 (315) | 0.190 (315) | 0.142 (312) | 0.166 (309) | |
| Nmass | <<0.001 (330) | 0.003 (330) | 0.012 (330) | 0.015 (330) | 0.015 (327) |
- Note: See Table 1 for meanings of abbreviations
| Tree | Intercept | Slope | λ | n | P value (slope) | Likelihood |
|---|---|---|---|---|---|---|
| APG1 BL | 0.396 | 0.475 | 1 | 96 | 0.0005 | 21.372 |
| APG1 BL | 0.577 | 0.233 | 0.8409 | 96 | 0.0572 | 44.314 |
| APG1 no BL | 0.545 | 0.239 | 1 | 96 | 0.0352 | 41.598 |
| APG1 no BL | 0.638 | 0.181 | 0.8191 | 96 | 0.1207 | 44.397 |
| Davies et al. | 0.541 | 0.226 | 1 | 94 | 0.0472 | 41.830 |
| Davies et al. | 0.653 | 0.151 | 0.8157 | 94 | 0.2032 | 44.183 |
| APG3 | 0.573 | 0.248 | 1 | 94 | 0.0307 | 41.146 |
| APG3 | 0.688 | 0.160 | 0.8291 | 94 | 0.1812 | 43.786 |
| OLS (no tree) | 0.580 | 0.421 | 0 | 96 | <0.001 | 32.775 |
- Note: OLS = ordinary least squares regression. See Table 1 for meanings of other abbreviations.
Both major and minor vein patterns were significantly related to leaf economic traits according to results of standard tests, suggesting functional links between venation and leaf economic traits across angiosperms (Tables 3, 4, Figs. 2, 3). However, P values for all tests were reduced with the use of phylogenetic methods, and results of many tests were not significant, reflecting the phylogenetic signal in the data (Tables 3, 4).
Box plots of major vein patterns vs. leaf functions. Left column: primary vein type vs. (A) Amass (N = 468 species), (B) Nmass (431 species), and (C) LMA (467 species). Right column: secondary vein type vs. (D) LLS (361 species), (E) LMA (315 species), and (F) Nmass (333 species). Letters indicate values that were statistically similar among categories according to standard ANOVA with Tukey test. Figure abbreviations: Amass, maximum photosynthetic rate on a mass basis; LLS, leaf life span; LMA, leaf mass per area; Nmass, nitrogen concentration on a mass basis.
Scatter plot of Aarea on minor vein density (N = 96 species), with r2 value from ordinary least squares regression. Figure abbreviation: Aarea, maximum photosynthetic rate on an area basis.
With the use of standard ANOVA, primary vein type was highly significantly related to Amass, LMA, and Nmass. With phylogenetic ANOVA, none of these relations was significant, regardless of which tree was used, so they do not appear to reflect correlated evolutionary change (Table 3). In contrast to my predictions, palmate- and parallel-veined leaves did not have consistently similar values of leaf economic traits. Palmate- and parallel-veined leaves both had significantly higher Amass than pinnate-veined leaves (Fig. 2A), but parallel- and pinnate-veined leaves both had higher Nmass and lower LMA than palmate-veined leaves (Fig. 2B, C). Consistent with predictions, secondary vein type was significantly related to LLS, LMA, and Nmass with both standard and phylogenetic ANOVA (Table 3, Fig. 2D–F). The difference between closed and intermediate secondary venation was not significant for any of the response variables; both closed and intermediate secondary veins had significantly longer LLS, higher LMA, and lower Nmass than did open secondary veins (Fig. 2D–F). Results of the highly significant χ2 tests indicated that the proportion of species with different major vein types was not distributed randomly among biomes (primary vein type × biome: χ2 = 47.6851, df = 16, P << 0.001; secondary vein type × biome: χ2 = 97.5017, df = 14, P << 0.001; Table 5).
| Primary vein type | Secondary vein type | |||||||
|---|---|---|---|---|---|---|---|---|
| Biome | Pinnate | Palmate | Parallel | Total | Closed | Intermediate | Open | Total |
| Alpine | 15 | 7 | 0 | 22 | 0 | 1 | 18 | 19 |
| Boreal | — | — | — | — | 0 | 0 | 2 | 2 |
| Desert | 2 | 1 | 0 | 3 | — | — | — | — |
| Grassland/meadow | 18 | 3 | 11 | 32 | 2 | 4 | 10 | 16 |
| Temperate forest | 158 | 15 | 17 | 190 | 24 | 48 | 56 | 128 |
| Temperate rain forest | 11 | 0 | 1 | 12 | — | — | — | — |
| Tropical forest | 19 | 0 | 5 | 24 | 14 | 18 | 5 | 37 |
| Tropical rainforest | 41 | 5 | 3 | 49 | 43 | 28 | 5 | 76 |
| Tundra | 13 | 2 | 2 | 17 | 4 | 9 | 15 | 28 |
| Wetland | 101 | 9 | 9 | 119 | 11 | 21 | 23 | 55 |
| Total | 378 | 42 | 48 | 98 | 129 | 134 | ||
A significant positive relation was found between MVD and Aarea with GLS regression with λ set to 0, which is equivalent to OLS regression (P < 0.001; Fig. 3, Table 4), but the explanatory power was quite low (r2 = 0.11). With the use of PGLS regression with λ set to its ML value, P values varied from 0.057 to 0.20, depending on the tree (Table 4). Regression models with λ = 1 (equivalent to independent contrast) had significant P values but also had significantly lower likelihoods (Table 4), suggesting that the models with the ML value of λ were more appropriate. For those species for which I had multiple leaves, within-species variation in MVD was much lower than among species variation, as shown by the significant effect of species nested within genus in an ANOVA of MVD (Appendix S13).
DISCUSSION
This study provides the first global-scale demonstration of relations between major vein patterns and leaf functions as well as the first phylogenetically based test of the relation between MVD and photosynthetic capacity. Significant relations, such as between secondary vein pattern and leaf economic traits, suggest selection for suites of traits is correlated with different vein patterns. Other relations, such as between primary vein type and leaf economic traits, were not significant according to phylogenetic tests, and they appear to be based on the conservatism of trait combinations in relatively few clades rather than mainly due to correlated evolution. The relation between MVD and Amax was marginally nonsignificant with phylogenetic regression, which, given the various sources of error in MVD and Amax measurements, does not rule out correlated evolution of hydraulic supply and demand. The presence of angiosperm-wide phylogenetic signal in every trait in this study emphasizes the role of phylogenetic conservatism in leaf functional trait relations. The results of this study have important implications for ecologists studying contemporary and paleo-ecosystems, both for postulating links between leaf form and ecosystem function and for predicting leaf function from leaf structure. They are also important for understanding how evolution generates leaf diversity, both in terms of the evolution of integrated functions such as hydraulic supply and demand (61) and the trade-off between photosynthetic capacity and support (77; 65).
Significance of leaf vein patterns for ecological studies
This study identified new links between major vein patterns and the LES and provided additional empirical support for the relation between MVD and leaf-level fluxes. Linking the LES to leaf structural traits such as vein patterns complements studies that have examined the chemical or genetic determinants of leaf economic traits (6; 49) and provides a basis for analyzing the ecological significance of leaf form diversity (59). The trade-off established by the LES has been used to understand processes ranging from species invasions (30; 39) to early angiosperm evolution (57) to community assembly (43; 1). The results presented here suggest that leaf vein architecture is an important correlate of a species’ position on the LES and may play a role in any of these ecological processes.
A link between vein patterns and the LES may be especially useful for analyses of plant fossil data. Vein patterns are highly visible in many fossils and do not require preservation of the entire leaf. In conjunction with other methods, vein characteristics could be used to estimate leaf economic traits for individual fossil taxa or assemblages of leaves (58; 11; 57). Although phylogenetic conservatism may weaken statistical relations between vein patterns and leaf functions, phylogenetic relations also can provide a complementary source of information for inferring unmeasured values of leaf traits with the use of regression techniques (27; 23; 14).
Significant differences in major vein patterns across biomes (Table 5) suggest that the leaf forms associated with different major vein patterns are important for adaptation to large-scale environmental variation. For primary vein types, the most notable deviation from the expected distribution was the abundance of parallel-veined leaves in grasslands, as might be expected, since all grasses had parallel veins. For secondary vein types, the most striking pattern is that closed-veined leaves were the most abundant form in tropical rainforests, whereas open-veined leaves were the most abundant in every other habitat except tropical forests, in which intermediate-veined leaves were the most common. These results should be considered tentative because no effort was made to sample evenly across biomes, and some biomes are very poorly represented. However, the results do warrant further study, for example, to determine whether relations between vein pattern and biome are due to the predominance of different clades in different habitats.
Minor vein density and the coordinated evolution of hydraulic capacity and photosynthetic capacity
The correlations between MVD and leaf-level fluxes of water or CO2 that were found in earlier studies (60; 12; 10; 11) suggest coordinated evolution of hydraulic supply capacity and photosynthetic capacity. I also found a highly significant relation between MVD and Aarea using standard regression, supporting a mechanistic link between the two traits, but when the phylogenetic signal in both traits was taken into consideration, the relation was reduced to marginally significant (Table 4). Because I used leaves from different sites to measure Aarea and MVD, a significant part of the unexplained variation may come from intraspecific variation, both genetic and plastic. Environmental conditions such as light exposure or nutrient availability can have a significant impact on both MVD and Amax, even within one site, and intrinsic factors such as leaf thickness or stomatal density also add scatter to the relation between Amax and MVD. Therefore, P values close to 0.05 (Table 4) make it difficult to reject the hypothesis that hydraulic capacity and photosynthetic capacity are evolving in a coordinated fashion. The lack of replication within species prevents me from statistically separating the effect of intraspecific variation from measurement error, but I used these measurements because I wanted to examine species-level relations, regardless of within-species variation. Despite the low r2 value, within-species variation in MVD was much less than among-species variation for those species for which I had multiple leaves (Appendices S13, S14), suggesting a strong species-level signal in MVD. Given the many sources of error in the measurements, even a weak signal in the data suggests both functional and evolutionary links between MVD and Amax. My results indicate that it will be difficult to precisely predict leaf-level physiology from MVD using unmatched samples, such as when estimating values of physiological traits for fossil leaves, but that the general trend of increasing Amax with increasing MVD is still valid across sites.
Major vein patterns and leaf-level trade-offs
The correlations of the LES are based on a trade-off between high photosynthetic capacity, facilitated by high N and water content, and long LLS, provided by traits like leaf toughness and physical support (77). The role of major veins in this trade-off is difficult to predict because investment in veins means less area and biomass for photosynthesis but also greater water delivery capacity, which can support higher rates of gas exchange. My results suggest that an evolutionary trade-off exists between photosynthetic capacity and investment in secondary veins, since species with open secondary veins had shorter LLS, lower LMA, and higher Nmass than those with closed and intermediate secondary veins (Fig. 2B). However, since I did not directly measure the proportion of leaf mass in secondary veins, the trade-off may be based on other traits that are correlated with LMA, such as smaller cells or a lower fresh mass/dry mass ratio (64; 65; 49). My results indicate no trade-off between photosynthetic capacity and the extra support offered by palmate or parallel primary veins, since leaves with palmate or parallel venation have higher Amass (Fig. 2A). Earlier work suggested that this relation is due to a trade-off between biomass allocated to primary veins and biomass allocated to structural tissue outside the primary veins (45), and although this may be true for palmate- vs. pinnate-veined leaves, it does not appear to be the case for parallel- vs. pinnate-veined leaves. Higher Nmass and lower LMA were associated only with palmate-veined leaves, whereas parallel- and pinnate-veined leaves were similar for those traits, despite differences in primary vein density. Furthermore, these relations do not appear to be based on an evolutionary trade-off, since they were not significant according to phylogenetic ANOVA.
High major vein density may be more important for the vascular redundancy it provides than for its role in support functions. Among primary vein types, the nonphylogenetic ANOVA indicates that higher Amass, higher Nmass, and lower LMA are associated with palmate venation, supporting the hypothesis of 59 that the vascular redundancy conferred by palmate venation should be most common in leaves with thin, vulnerable laminas. However, the nonsignificant results for the phylogenetic ANOVA suggest that this relation did not arise through multiple instances of correlated trait evolution. Parallel-veined leaves, which I predicted would have similar vascular redundancy to palmate-veined leaves, grouped inconsistently with either pinnate- or palmate-veined leaves (Fig. 2A–C), suggesting that vascular redundancy is not the primary factor selecting for parallel venation. Among secondary vein types, intermediate-veined leaves were statistically similar to closed-veined leaves (Fig. 2D–F), suggesting that any putative differences in vascular connectivity are not reflected in leaf function. This is consistent with the findings of 59 that severing secondary or higher-order veins had no effect on leaf hydraulic function, and it suggests that the extra secondary vein tissue at the leaf margin is more important for support than for hydraulic redundancy. Future empirical studies that use leaves of different major vein types will help to clarify any differences in function among vein types and shed light on the relative importance of hydraulic vs. support functions in optimizing major vein patterns.
Phylogenetic conservatism and leaf functional trait relations
This study provides evidence that leaf traits are phylogenetically conserved and that phylogenetic conservatism can have a significant impact on leaf form–function relations. Phylogenetic conservatism of major vein patterns has been recognized for some time, as summarized in 59. In this study, primary vein pattern was invariable for most of the families surveyed (Appendix S15). With the use of unordered parsimony reconstruction, 39 transitions in primary vein type were found among the 468 species scored (Appendix S16). Secondary vein type was more labile than primary vein type, with 98 to 124 transitions (depending on the tree used) among 361 species, but few families had both open and closed secondary veins (Appendices S16, S17). Although phylogenetic ANOVA is analogous to reducing N from the number of species to the number of transitions, thereby reducing the power of the ANOVA, there were still 39 transitions in primary vein type, which should provide a sufficiently large sample to detect the correlated evolution of primary vein type and leaf economic traits. The association between the lability of vein types and the significance of phylogenetically based relations (at least for these two samples) suggests that the relative evolutionary lability of a trait is linked to the likelihood of its evolving in concert with other traits.
Phylogenetic conservatism of primary vein type suggests that its relations with leaf economic traits have arisen from specific combinations of traits found within a few clades. For example, 22 of the 48 parallel-veined species were grasses (Poaceae). Excluding grasses from the analysis had a significant impact on the results of a standard ANOVA testing for differences in Amass (pinnate- and parallel-veined leaves had similar Amass, whereas palmate- and parallel-veined leaves were statistically different; compare with Fig. 2A). Despite the prevalence of C4 photosynthesis in parallel-veined species (Appendix S15), excluding C4 species from the analysis had no significant impact on the results, suggesting that the results are not driven by the associations between C4 photosynthesis and parallel venation and that other traits should be investigated.
Even MVD, which is generally considered to be quite labile (55), had significant phylogenetic signal at this scale (Table 1). This result is consistent with a recent study that showed differences in MVD among large clades of land plants (10). It is an open question whether the phylogenetic patterns in MVD are a result of developmental constraints similar to those found in major vein patterns (19) or of exogenous factors such as stabilizing selection acting similarly among close relatives (e.g., phylogenetic niche conservatism; 74). The presence of angiosperm-wide phylogenetic signal in leaf economic traits was also novel. Although earlier work is suggestive of phylogenetic signal in leaf economic traits (2; 24), studies among close relatives have demonstrated that photosynthetic traits can evolve rapidly (4; 28). Because the phylogenies in this study are only resolved to the family, or sometimes genus, level, these results are not necessarily inconsistent. Phylogenetic signal may vary across scales, and the scale of the study may determine its importance (15; 50).
The results of this study may be sensitive to many variables, including the phylogenetic hypothesis, taxon sampling, and the model of evolution. The results across four trees suggest that the precise form of the backbone phylogeny does not have much effect, at least for the phylogenies currently available. Taxon sampling may affect the choice of the most likely evolutionary model (Table 1), and although it does not appear to affect the overall results or conclusions in this study, future studies with more taxa or taxa that are more evenly distributed across the tree could give different results. The choice of evolutionary models does not change the conclusions of any of the ANOVAs, but it does change the regression of Aarea on MVD from marginally nonsignificant to significant (compare models with λ set to 1, vs. λ set to its ML value, Table 4). This demonstrates the importance of considering alternatives to the Brownian motion model in phylogenetic comparative analyses. Better resolution of the phylogeny also could affect the results, particularly for tests of phylogenetic signal. However, better resolution is unlikely to change the results of the ANOVA for primary vein type, since it is largely invariable within families. It is also unlikely to change the results of the phylogenetic ANOVAs for secondary vein types, since these results are already significant. Including branch lengths had an effect on the most likely evolutionary model for several traits (Tables 2, 3), which, given the uncertainty in branch lengths, provides a potentially large source of error.
Conclusions
Leaf traits analyses form the basis for understanding how differences in leaf form affect organismal, community, and ecosystem function, yet few broad-scale phylogenetic analyses of leaf trait relations exist. In this study, I used both standard and phylogenetic methods to demonstrate global-scale relations between vein patterns and leaf functions as well as the effects of phylogenetic conservatism on leaf trait relations. The phylogenetic comparative methods used here provide a way to test alternative hypotheses of trait evolution and offer an emerging framework for the study of plant physiology. Although these methods do not speak directly to the mechanistic causes or ecological consequences of leaf trait relations, they can be used to generate new mechanistic or ecological hypotheses. The prevalence of phylogenetic conservatism in leaf traits may not prevent correlated evolution, but it may affect the course of evolution. Genetic, developmental, or other constraints preventing transitions in traits such as major vein type may force species of a particular clade to evolve alternative leaf characteristics to compensate, leading to much of the scatter in Figs. 2 and 3. In this way, conserved traits may force the evolution of alternative phenotypes, promoting phenotypic diversification rather than impeding it and helping to explain the high diversity of angiosperm leaf forms.
