Significance Leaf traits are used to drive models of global carbon fluxes and understand plant evolution. Many syntheses have highlighted relationships between plant leaf nitrogen and photosynthesis as evidence of a strong evolutionary drive to “intercept light and capture CO 2 .” Different from previous studies, we compiled a global dataset constrained to sites and studies where nitrogen-fixing plants (N 2 FP) and nonfixing species [other plants (OP)] could be directly compared. We show that photosynthesis is not related to leaf nitrogen for N 2 FP, irrespective of climate or growth form. N 2 FP have clear advantages in water use efficiency over OP. These findings contribute to a more complete explanation of global distributions of N 2 FP and can help improve models of global carbon and nitrogen cycles.

Abstract Using robust, pairwise comparisons and a global dataset, we show that nitrogen concentration per unit leaf mass for nitrogen-fixing plants (N 2 FP; mainly legumes plus some actinorhizal species) in nonagricultural ecosystems is universally greater (43–100%) than that for other plants (OP). This difference is maintained across Koppen climate zones and growth forms and strongest in the wet tropics and within deciduous angiosperms. N 2 FP mostly show a similar advantage over OP in nitrogen per leaf area (N area ), even in arid climates, despite diazotrophy being sensitive to drought. We also show that, for most N 2 FP, carbon fixation by photosynthesis (A sat ) and stomatal conductance (g s ) are not related to N area —in distinct challenge to current theories that place the leaf nitrogen–A sat relationship at the center of explanations of plant fitness and competitive ability. Among N 2 FP, only forbs displayed an N area –g s relationship similar to that for OP, whereas intrinsic water use efficiency (WUE i ; A sat /g s ) was positively related to N area for woody N 2 FP. Enhanced foliar nitrogen (relative to OP) contributes strongly to other evolutionarily advantageous attributes of legumes, such as seed nitrogen and herbivore defense. These alternate explanations of clear differences in leaf N between N 2 FP and OP have significant implications (e.g., for global models of carbon fluxes based on relationships between leaf N and A sat ). Combined, greater WUE and leaf nitrogen—in a variety of forms—enhance fitness and survival of genomes of N 2 FP, particularly in arid and semiarid climates.

Through symbioses with diazotrophic bacteria, legumes and other N 2 -fixing plants (N 2 FP) acquire atmospheric dinitrogen (N 2 ) and are widely expected to maintain greater leaf nitrogen than nonfixing or other plants (OP) (1). N 2 FP can profoundly influence both ecosystem development and responses to changing climate by alleviating nitrogen shortages that limit capacity of ecosystems to fix and sequester CO 2 (2⇓–4). A central tenet of trait-based ecology (5, 6) is that carbon fixation and transpiration are directly related to leaf nitrogen; in turn, leaf nitrogen is used to drive global models of carbon (and water) exchanges between plants and the atmosphere (7).

The distribution, abundance, and activity of N 2 FP in terrestrial ecosystems have remained unexplained, even “paradoxical” (8, 9), especially in relation to local and global nitrogen cycles. For the northern hemisphere, one recent explanation of the distribution of N 2 FP (2) and their dominance in wet tropical forests relied on their greater ability to acquire phosphorus from old tropical soils and temperature maxima for N 2 fixation of around 25 °C (i.e., similar to prevailing temperatures in the tropics). Menge et al. (8) subsequently noted that the diazotrophic symbioses are typically rhizobial and facultative toward the tropics but actinorhizal and obligate north of about 35° N. Facultative symbioses in the tropics make evolutionary sense inasmuch as soil nitrogen availability is typically greater there than at the poles and nitrogen fixation carries a carbon cost for the plant. In support, concurrent research suggested that rates of nitrogen fixation may be less in N-rich tropical forests than previously thought (10).

N 2 FP differ in their distribution in northern and southern hemispheres, albeit that N 2 FP are common in the tropics in both hemispheres. By comparison with the north, beyond 35° S, there is relatively little land at all. Bryophyte–cyanobacteria associations again contribute significant nitrogen (11), albeit to much smaller areas than in the northern hemisphere, and actinorhizal plants (e.g., Morella/Myrica spp. in Africa and South America and Casuarina spp. in Australia) are as likely found in the tropics as closer to the southern pole (12). A distinctive feature of all three major continents in the southern hemisphere is the large areas of arid, semiarid, and Mediterranean (summer drought) climates between the equator and 35° S. In divergence from the “view from the north” (13), the “southern paradox” of the distribution of N 2 FP is that woody legumes, notably of the genus Acacia (sensu lato) but also, from numerous other genera, dominate much of the large arid and semiarid areas, despite an abundance of other drought-tolerant woody species. For Australia, the paradox is exemplified by the dominance of Acacia aneura and Acacia harpophylla over large areas, whereas nominally drought-adapted species from the genus Eucalyptus are restricted to drainage lines or where groundwater is accessible.

Analysis of plant traits is now routinely used (14⇓⇓⇓–18) to seek explanations for distributions of plant species and growth forms as well as their functional attributes. Leaf nitrogen is among the most significant and widely explored of plant traits. For example, it is frequently observed that leaf nitrogen is greater per unit mass or area for N 2 FP than for OP (1). Leaf nitrogen has been a focus for trait-based studies of plants owing in part to strong positive relationships between leaf nitrogen and photosynthetic rate (19) and the implications for stomatal conductance (g s ) and transpiration (20, 21). Increased leaf nitrogen (especially increased abundance of the principal nitrogen-rich enzyme involved in carbon fixation; RubisCo) can increase consumption of intercellular CO 2 , such that g s is reduced (and rates of water loss are reduced), because a strengthened CO 2 diffusion gradient helps maintain supply of CO 2 . A corollary is that maintaining rates of photosynthesis (A sat ) with reduced leaf nitrogen may require increased g s and water loss. Recently, Prentice et al. (22) built on earlier analysis by Wright et al. (5) and proposed a new theoretical framework for plant ecology based on leaf traits, such as nitrogen per leaf area (N area ), A sat , g s , and the ratio of internal to external concentration of carbon dioxide (c i /c a ). Prentice et al. (22) focused on the relative constancy of c i /c a over a wide range of conditions, tested their theory using sites in Australia, including Acacia spp. and other N 2 FP, and argued that N area should increase with aridity and that high N area is an adaptation to drought. Despite some recent studies (23), that theory lacks testing for N 2 FP across the globe.

To test “paradoxes” associated with the global distribution of N 2 FP, we formalized hypotheses in accordance with the literature. Leaf nitrogen should reflect rates of A sat (hypothesis A)—irrespective of whether the plant species can fix nitrogen. Increases in leaf N should, thus, result in reduced g s and loss of water (hypothesis B) and as a result of either or both, increase water use efficiency [WUE; as indicated by intrinsic water use efficiency (WUE i ) or carbon isotope ratio of leaf tissue (δ13C); hypothesis C].

We tested our hypotheses using a climate-stratified dataset constrained to sites where both N 2 FP and OP (paired dataset) were measured for either (i) N area , A sat , g s , and WUE i or (ii) N area and δ13C (that is, sites where N 2 FP and OP were both growing and measured in situ). We complemented this parsimonious, albeit more limited dataset (81 sites) with a larger dataset, in which either N 2 FP or OP were studied (nonpaired dataset) for WUE i (including A sat and g s ) and N area (63 sites) or δ13C and nitrogen concentration per unit leaf mass (N mass ; 351 sites). We adopted the Koppen system—the most frequently used and robust method for climate classification and related analyses (24, 25).

Discussion Positive relationships between leaf N and A sat have been widely reported at scales ranging from individual plant species to the globe. For example, our independent analysis for OP (Fig. 3B) is qualitatively similar to those in the works by Evans (19) and Wright et al. (5). However, our analysis also shows that this is not the case for N 2 FP in nonagricultural ecosystems (Fig. 3A), and the literature shows that it is not true for agricultural systems (26). Our results also challenge the prevailing theory that additional leaf N will increase A sat or reduce g s (20). We found that additional leaf N was only ever a positive influence on both A sat and g s . Osnas et al. (6) and many others draw on the broad observation that leaves have evolved primarily to intercept light and capture CO 2 to propose that photosynthetic capabilities are mostly proportional to leaf area. There are, however, other evolutionary forces at work. Given the lack of support among N 2 FP for either greater carbon gain (hypothesis A) or reduced leaf water loss (hypothesis B) but good evidence for enhanced WUE (hypothesis C), can these other forces help explain leaf N and the dominance of many arid and semiarid zones by woody legumes? Rates of leaf and plant growth are only part of evolutionary success and must be considered alongside a plant’s ability to survive and reproduce. Relative to photosynthetic needs, overinvestment of nitrogen in leaves in harsh semiarid to arid regions has remained unexplained (22). In these areas, there is little selection pressure for light, to create a large canopy, or to grow quickly. A potent selective force is the ability to survive (as either plant or seed) periods of drought that might last weeks to months or even a decade or more. For annual agricultural legumes, Hardwick (27) noted that canopy A sat varies according to the rate of growth of the seed—not the other way around. There is also abundant evidence that remobilization of nitrogen from foliage and other plant tissues may account for 70–90% of seed nitrogen in annual agricultural legumes (28). Prolific flowering and generation of seedpods and seeds are features of many N 2 FP (Fig. S1). Although it is not known how much nitrogen is remobilized from leaves to seeds for the thousands of species of N 2 FP in nonagricultural ecosystems, current knowledge suggests that leaf N is an investment in the ability of N 2 FP to produce seed and the “survival of the genome” (27). Furthermore, the competitive ability of N 2 FP is enhanced by their ability to take up other forms of N available in the soil (29) or when diazotrophy is restricted by water availability (30, 31). N 2 FP also make efficient use of N temporarily stored in foliage. For example, in the forms of amines, polyamines, alkaloids, cyanogenic glucosides, and many others, N-rich molecules help N 2 FP cope with drought (by osmotic adjustment) as well as freezing conditions (32) and also, help deter herbivores in both tropical and nontropical forests (33, 34). Fig. S1. Flowering woody legumes from Africa and Australia. (Upper Left) Ormocarpum tricocarpum in Kruger National Park, South Africa. (Upper Right) Acacia boormanii in southeast Australia. (Lower Left) Acacia macradenia in northern Australia. (Lower Left) Swainsonia Formosa in northwestern Australia. Despite relatively recent evolution (∼60 MyBP) (35, 36), possibly from a “single cryptic evolutionary innovation” (36), symbioses with diazotrophic bacteria ensure access of N 2 FP to nitrogen—one of the most limiting resources for plant growth, survival, and reproduction. That insurance and other nitrogen-related advantages have facilitated the spread of N 2 FP throughout the globe and their contributions to global N cycles (37, 38). The facultative nature of the symbiosis with respect to soil nitrogen (4, 8⇓–10) is augmented by its flexibility in relation to soil water—N 2 FP seldom fix nitrogen under drought conditions (29⇓–31), although their ability to nodulate may be unimpeded (39) and help restore fixation after drought is relieved. These features facilitate the dominant role played by N 2 FP in both wet and dry tropics as well as large areas of temperate and Mediterranean climates. WUE contributes further to the evolutionary advantages enjoyed by legumes and other N 2 FP. In their recent synthesis of the now large body of work that informs our understanding of δ-values in plants, Cernusak et al. (21) noted that, for C 3 plants, the range in δ-values (Cernusak used Δ in place of δ) was constrained by coordination of g s and A sat . A more sophisticated and complex relationship between δ and WUE than what was once recognized does not detract from the evidence presented here that the latter contributes to our knowledge of the benefits enjoyed by legumes and why they are different from OP (40). If trait-based models of regional and global carbon cycles (7) are to achieve promised predictive capabilities, they will need to incorporate WUE as well as traits, such as the ability of N 2 FP to store and use N in leaves for other survival-related functions. Increasingly dry conditions in many areas of the globe reinforce this point. In similar fashion, the absence of significant predictive power of leaf nitrogen for rates of carbon fixation by N 2 FP will pose ongoing challenges given their dominance of so many wet tropical forests that collectively are critical to global C cycles.

Methods Data Acquisition. We developed a database from a global meta-analysis of published literature (Table S4). Our database was targeted to our hypotheses; studies included from natural systems had to contain a measure of leaf nitrogen content and a measure of leaf WUE for N 2 FP and OP. We identified relevant literature by screening the Web of Science and Google Scholar search engines for keywords: carbon isotope discrimination, 13C, WUE, water use efficiency, leaf nitrogen, legume*, n-fix*, and nodulation; it also included relevant citations documented in these literature. We included targeted searches for each of the major actinorhizal genera. Table S4. List of references from which data were drawn We constructed two datasets: one based on studies with concurrent data that were collected from the same site for both N 2 FP and OP (paired dataset) and one that included studies with data for either N 2 FP or OP presented (nonpaired dataset). For each of the paired and nonpaired datasets, we had two subsets: one comprised of data of N mass (milligrams gram−1) and δ13C (percentage) recorded concurrently and one comprised of data for studies of N area (grams meter−2) reported concurrently with WUE i (micromoles CO 2 moles−1 H 2 O) or both A sat (micromoles CO 2 meter−2 second−1) and g s to water vapor (moles meter−2 second−1), such that we could calculate WUE i . The paired dataset includes 22 sites across the globe for studies that presented data in a form from which we could record or calculate N area together with WUE i and 81 sites containing data in a form from which we could record or calculate N mass and δ13C, with 57 of those sites also presenting data for specific leaf area (meters2 kilogram−1) or leaf mass per unit area (grams centimeter−2), which enabled calculation of N area . The nonpaired dataset contains 63 sites across the globe for N area and WUE i and 351 sites for N mass and δ13C. For studies where a treatment was applied, only data from the control were used. Species were identified as N 2 FP (including actinorhizal and nodulating plants) or OP and classified by their growth form: fern, forb, graminoid, gymnosperm, woody evergreen angiosperm, or woody deciduous angiosperm. In total, 11 actinorhizal species were included, the majority of which are from the families Rosaceae or Casuarinaceae (Fig. S2). Digital latitude and longitude of each site were recorded and used to identify site mean annual temperature (degrees Celsius), mean annual precipitation (millimeters), dryness index (mean annual precipitation/potential evaporation), and elevation (meters a.s.l.). We also identified sites according to their Koppen classification (A, tropical/megathermal; B, dry/arid/semiarid; C, temperate/mesothermal; and D, continental/microthermal). Fig. S2. Growth forms represented in each Koppen climate classification [A (tropical), B (dry, arid, and semiarid), C (temperate) and D (continental)] for N 2 FP (red bars) and OP (blue bars) across (A) the paired δ13C dataset and (B) the WUE i dataset. Numbers above the bars represent numbers of observations, with the numbers in parentheses representing actinorhizal species. Data Analysis. Shapiro–Wilk tests showed that data for all variables were significantly nonnormal (skewed to the right). Log 10 transformations improved normality distributions of data for all variables except δ13C, which had distribution that did not improve with either log 10 or square root transformation; hence, all analyses were performed on nontransformed δ13C data. We used multivariate analyses (linear mixed models and maximum likelihood) to quantify the combined influence of N-fixing status, climate variables, and growth form on leaf nitrogen. Site and author were treated as random factors for all analyses to counter nonindependence. We used bivariate analyses (Pearson correlations) to assess simple relationships between measures of WUE and leaf nitrogen content or measures of WUE and climate-related variables. Multivariate stepwise multiple regressions better explained relationships in toto among leaf nitrogen, climate, and leaf WUE. The large range in data for bivariate analyses was conserved between N 2 FP and OP groups. All analyses were performed with SPSS. Unless denoted otherwise, data and analyses refer to the paired dataset.

Acknowledgments We thank Alexandra Barlow for helping us screen the literature. We also thank the numerous authors who provided additional data on request and the two reviewers for their suggestions that significantly improved this article. We thank the Australian Research Council for support. ETH Zurich is thanked for its support to M.A.A. as a visiting professor.

Footnotes Author contributions: M.A.A. and T.L.T. designed research; M.A.A. and T.L.T. performed research; M.A.A., T.L.T., J.I.S., and N.B. analyzed data; and M.A.A., T.L.T., J.I.S., and N.B. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1523936113/-/DCSupplemental.

Freely available online through the PNAS open access option.