We analysed 2228 leaf samples from 670 species of terrestrial plants, representing 138 families. Raw data for all samples are presented in a database in the Supplementary material, Table S1. Angiosperm and gymnosperm family assignments were based on the Angiosperm Phylogeny Website (http://www.mobot.org/MOBOT/Research/APweb/welcome.html). Genera of ferns and their allies follow the classification of the Australian National Herbarium (http://www.anbg.gov.au/fern/taxa/classification.html). Species names were checked against the International Plant Names Index (http://www.ipni.org/index.html). All species were assigned to 30 ‘key clades’, which represent monophyletic groups at relatively high taxonomic levels following recent phylogenetic insights (Pryer et al., 2001; Soltis et al., 2005), including Bryophytes, Lycopodiophytina, Polypodiophytina, Equisetophytina, Ginkgoales, Pinales, Cycadales, Austrobaileyales, Alismatales, Dioscoreales + Pandanales, Liliales, Asparagales, Commelinids, Magnoliids, Ranunculales, Sabiaceae, Proteales, Berberidopsidales, Caryophyllales, Santalales, Saxifragales, Vitales, Crossosomatales, Myrtales, Eurosids I, Eurosids II, Cornales, Ericales, Euasterids I, and Euasterids II (Fig. 1).
Samples were collected using statistically unstructured sampling techniques, from 1980 to 1984, at 26 sites in Japan (Aomori, Gifu, Gunma, Hokkaido, Hyogo, Ibaraki, Ishikawa, Kagoshima, Kanagawa, Kochi, Kyoto, Miyagi, Nagano, Nagasaki, Nara, Niigata, Okayama, Okinawa, Osaka, Saitama, Shiga, Tochigi, Tokushima, Tokyo, Tokyo-Ogasawara, Tottori) and at a further three sites in Indonesia, New Zealand and Sweden. The underlying soils, topography and climate (from cool-temperate to subtropic) of these sites varied substantially (data not available). Newly emerged and perennial leaves were collected randomly, rinsed in deionized water, and dried at 60°C. Leaf veins were removed, and the remaining sample was pulverized (Cyclotec 1093 Sample Mill; Foss Tecator, Höganäs, Sweden). Subsequently, 100 mg of sample was analysed by neutron-activation analysis (Koyama & Matsushita, 1980). Samples were packed in double polyethylene film bags with a neutron flux monitor. The bags were irradiated (Kyoto University Nuclear Reactor, Kyoto, Japan). For short-lived nuclides, samples were irradiated in a pneumatic transfer tube (Pn-3, thermal neutron flux; Φth = 2.3 × 1013 n cm−2 s−1) for 60 s. After sufficient cooling, gamma-ray spectra were measured for 200 s using a diode detector system of 90cc-Ge(Li) coupled to a 4096-channel pulse-height analyser. For longer-lived nuclides, samples were irradiated in a pneumatic transfer tube (Pn-2, thermal neutron flux; Φth = 2.8 × 1013 n cm−2 s−1) for 1 h. For intermediate- and long-lived nuclides, samples were cooled (164 h and 1 h, respectively), and gamma-ray spectra measured (1 h). Forty-four elements could be analysed using this method: Ag, Al, As, Au, Ba, Br, Ca, Cd, Ce, Cl, Co, Cr, Cs, Cu, Dy, Eu, Fe, Gd, Hf, Hg, I, K, La, Lu, Mg, Mn, Mo, Na, Nd, Ni, Rb, Sb, Sc, Se, Sm, Sr, Ta, Tb, Th, Ti, U, V, Yb and Zn. Note, despite splitting samples to analyse elements with different half-lives, it was still not possible to obtain data for all elements for all samples because of variation in half-lives among groups and overlap of Compton peaks. Raw leaf concentration data for 42 elements were loge-transformed (data available in a database, which is published as supplementary information on the New Phytologist website). Both I and Ti were excluded from the analysis because of sample sizes of n ≤ 10.
A variance components model suited to analysing unstructured data was used to allocate variation in leaf mineral composition to a phylogenetic component defined as ‘(key clade/family/species)’, using residual maximum likelihood (REML) procedures (Broadley et al., 2001, 2003, 2004, 2007; White et al., 2004; Hodson et al., 2005). The overall random term within the variance components model was (site + (key clade/family/species)) and no fixed factors were defined. Thus, variation in leaf mineral composition caused by soil chemistry and/or climate was assigned to the ‘site’ component of the model. Variation in leaf mineral composition resulting from sampling (e.g. leaf age, sampling height, etc.) was assigned to the residual term. Therefore, the following questions can be addressed: how big is the phylogenetic effect, compared with the site + residual effects, how much of this phylogenetic variation is at the species level and how much at the family level, etc.? Such techniques have previously been used to extract evolutionary information from other unstructured data-sets including data from the literature (Broadley et al., 2001, 2003, 2004, 2007; White et al., 2004; Hodson et al., 2005).