1 INTRODUCTION

An influential review (Field & Mooney, 1986) helped to establish the relationship between leaf nitrogen (N) and photosynthesis as a cornerstone of modern ecophysiological theory. Drawing on previous empirical studies and reviews, Field and Mooney (1986) considered the implications of observed correlations between the light-saturated photosynthetic rate (as a measure of photosynthetic capacity) and leaf N content, expressed on either an area or a mass basis. Numerous studies have confirmed the positive correlation between photosynthesis and leaf N at local, regional and global scales (see e.g. Evans, 1989; Niinemets, 1999; Wright et al., 2005). The correlation exists because photosynthetic processes depend on N-rich enzymes, and these—most importantly the primary carbon-fixing enzyme, ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco)—comprise a substantial proportion of leaf N.

A heuristic stemming from work on herbaceous crops (Evans, 1983; Evans & Seemann, 1989) is that photosynthetic proteins together account for 50%–60% of leaf N, of which Rubisco alone accounts for 25–30%. In non-crop and woody species, however, Rubisco often contributes a smaller percentage of total leaf N than in crops. A recent estimated of N partitioning in a ‘typical’ C₃ leaf put the Rubisco contribution at 20%, and total photosynthetic N (in light-absorbing pigments, electron-transport proteins and enzymes of the Calvin cycle including Rubisco) at 54% of leaf N (Evans & Clarke, 2018). Onoda et al. (2017) indicated that Rubisco contributed on average <15% of leaf N across 70 woody evergreen species. They estimated mean N in cell walls as 15% of leaf N in woody evergreens (as large a contribution as Rubisco), and 11% across all species. They also found that the proportion of leaf N allocated to cell walls increases with leaf dry mass per unit area (LMA), pointing to the importance of structural as well as metabolic components of leaf N.

The representation of N cycling, with implications for primary production and its response to rising atmospheric CO₂ and regional N deposition, has become a priority in the development of dynamic global vegetation models (DGVMs)—including both ‘offline’ and coupled DGVMs (Meyerholt et al., 2020; Zaehle & Dalmonech, 2011). However, current DGVMs show large differences in their modelled responses of primary production to carbon dioxide (CO₂) and N enhancement, reflecting a fundamental lack of agreement on the processes involved (Davies-Barnard et al., 2020). Key areas that are not well understood include the controls of gains and losses of reactive nitrogen at the ecosystem level, and the relative importance of nitrogen supply (from the soil, and/or N-fixing symbionts) and demand (by plants) in determining leaf N. Many coupled carbon–nitrogen (C-N) models can reproduce the CO₂-induced enhancement of tree growth in Free Air Carbon Enrichment (FACE) experiments but have been shown to do so via the wrong mechanism, that is, they rely on shifts in leaf-level stoichiometry to reduce N demand (Finzi et al., 2007; Medlyn et al., 2015; Zaehle et al., 2014) rather than increasing N uptake, which is the principal mechanism at the Duke Forest and Oak Ridge FACE sites (Medlyn et al., 2015; Norby et al., 2010; Zaehle et al., 2014). Dong, Wright, et al. (2022) quantified leaf-level N declines in response to CO₂ enrichment; however, increased NPP nonetheless creates increased N demand at the whole-plant level. Many C-N models rely on the hypothesis that photosynthetic capacity is determined by leaf N (Walker et al., 2014; Zaehle & Dalmonech, 2011), which, in turn, is determined by soil N availability (Luo et al., 2004). The causality of these relationships is open to question, however. They have not been extensively tested.

A perspective on plant functional traits based on eco-evolutionary optimality theory (Franklin et al., 2020; Harrison et al., 2021) suggests an alternative approach to the interpretation and modelling of leaf N, which we explore in this paper. We start from the observation that leaf N per unit leaf area (N_area) is related to the amount of leaf tissue (indexed by LMA) and its metabolic activity (indexed by carboxylation capacity at 25°C, known as V_cmax25). Then, we test the hypothesis that both LMA and V_cmax25 adjust to the growth environment through some combination of phenotypic acclimation, genetic adaptation and environmental selection (Dong et al., 2020; Kikuzawa et al., 2013; Smith et al., 2019; Wang et al., 2017), approaching predictable community-mean values. V_cmax is predicted here by combining the coordination hypothesis (Chen et al., 1993; Dong et al., 2017; Peng et al., 2021; Smith et al., 2019; Togashi et al., 2018)—that V_cmax at growth temperatures acclimates so that photosynthesis is (on average) neither limited by light nor by Rubisco activity—with the least-cost hypothesis—that stomatal behaviour minimizes the combined costs of maintaining both V_cmax and the water-transport pathway required to meet transpirational demand. Both costs can be expressed as functions of χ, the ratio of leaf internal to ambient CO₂, allowing the optimal χ to be predicted from environmental variables (Prentice et al., 2014). The value of χ, in turn, determines how large V_cmax must be, to satisfy the coordination hypothesis (Wang et al., 2017). LMA is predicted here via a model based on Kikuzawa's (1991) hypothesis that leaf life span maximizes lifetime-average net carbon gain after deduction of amortized construction costs, Wang et al. (2021) extended this hypothesis to a prediction of how the leaf economics spectrum (the relationship between LMA and leaf life span), and the expected distribution of LMA values, shift as a function of the growth environment.

Dong et al. (2017) showed that more than half of the variance in N_area across multiple plant types and environments in Australia could be accounted for by a linear combination of measured LMA and optimal V_cmax25 values as predicted via the coordination hypothesis. A similar approach has been used to predict observed trends in N_area, based on field measurements of both quantities, along regional transects in the Andes (South America) and Gongga (western China) mountains, respectively (Peng et al., 2020; Xu et al., 2021). Here we present an analysis of a large, newly compiled global leaf N dataset, with associated gas-exchange measurements for all species. Our analysis proceeded in steps as follows. First, we analysed how leaf N depends statistically on measured LMA and V_cmax25, thereby testing and quantifying Dong et al.’s (2017) hypothesis—that N_area can be well approximated as the sum of components proportional to V_cmax25 and LMA, respectively—at a global scale. Second, we analysed how variations in site-mean V_cmax25 and LMA depend statistically on climate, and compared these empirical dependencies with those independently predicted by leaf-level optimality theory. Combining these first two steps, we then tested the extent to which variations in site-mean N_area could be predicted from climate via independently predicted site-mean values of V_cmax25 and LMA. Noting the evidence for effects of soil fertility factors on optimal leaf traits (Paillassa et al., 2020), we also analysed the extent to which deviations from climate-based predictions for V_cmax25, LMA and N_area were linked to soil clay content, pH and C:N ratio as an inverse proxy for N availability.

Our analysis focuses principally on N_area because of its relationship to photosynthetic light absorption. However, we carried out additional statistical analyses of leaf N per unit mass (N_mass), which relates to tissue stoichiometry and (unlike N_area) is measured independently of LMA. Osnas et al. (2013) provided a theoretical analysis of the relative merits of area- and mass-based expressions of leaf nutrients and photosynthetic traits. They indicated that area-based expressions were generally more appropriate, but also that leaf N includes both an area-proportional and a mass-proportional component, consistent with our approach.