Relationships between soil and leaf mineral composition are element‐specific, environment‐dependent and geographically structured in the emerging model Arabidopsis halleri

Summary Leaf mineral composition, the leaf ionome, reflects the complex interaction between a plant and its environment including local soil composition, an influential factor that can limit species distribution and plant productivity. Here we addressed within‐species variation in plant–soil interactions and edaphic adaptation using Arabidopsis halleri, a well‐suited model species as a facultative metallophyte and metal hyperaccumulator. We conducted multi‐element analysis of 1972 paired leaf and soil samples from 165 European populations of A. halleri, at individual resolution to accommodate soil heterogeneity. Results were further confirmed under standardized conditions upon cultivation of 105 field‐collected genotypes on an artificially metal‐contaminated soil in growth chamber experiments. Soil‐independent between‐ and within‐population variation set apart leaf accumulation of zinc, cadmium and lead from all other nutrient and nonessential elements, concurring with differential hypothesized ecological roles in either biotic interaction or nutrition. For these metals, soil–leaf relationships were element‐specific, differed between metalliferous and nonmetalliferous soils and were geographically structured both in the field and under standardized growth conditions, implicating complex scenarios of recent ecological adaptation. Our study provides an example and a reference for future related work and will serve as a basis for the molecular–genetic dissection and ecological analysis of the observed phenotypic variation.


Fig. S1
Map of European sampling sites and edaphic range of A. halleri.        Fig. S5.

Table S4
Composition of Zn-and Cd-amended soil mix for plant cultivation under controlled growth chamber conditions. Methods S1. Detailed Methods. Plant and soil sampling in the field Locations of field sites hosting populations of Arabidopsis halleri (L.) O'Kane and Al-Shehbaz in Europe were assembled from the Global Biodiversity Information Facility (GBIF, http://www.gbif.org/), published records (Kolník & Marhold, 2006;Koch & Matschinger, 2007;Godé et al., 2012;Pauwels et al., 2012) and internet searches for historical or present Zn, Pb and Cu mines and smelters. At 165 field sites, we took one pair of a leaf and a soil sample from each sampled plant individual (Fig. S1, Notes S1; n = 3 to 20 individuals per site, averaging 12 individuals; minimum distance to sampled neighbour 3 m; 3 June to 29 October 2011, 30 July to 13 October 2012). Leaf samples (4 to 10 of the youngest undamaged fully expanded leaves) were washed thoroughly in deionized water. Soil samples (between 50 and 250 g) were taken using a stainless steel soil coring device (diameter 0.014 m) at 0.05 to 0.15 m depth within a 0.05 m radius around each plant individual. Samples were placed into paper bags and left to dry in ambient air.
Processing and analysis of samples After additional drying (60°C for ≥ 3 d, ambient air for ≥ 1 d) and homogenization (≤ 1 mm particle size by manually squeezing paper bags), a subsample of 10 to 25 mg leaf tissue was weighed into PTFE MPV-100 microwave vessels (MLS GmbH, Leutkirch, Germany), manually mixed with 3 ml 65% (w/w) HNO3 (AnalaR, Merck Ltd, Darmstadt, Germany), microwave-digested with temperature ramping to 180°C over 20 min and holding for 10 min (StarT-1500, MLS GmbH, Leutkirch, Germany), transferred into 15-ml round-bottom polypropylene screw-cap tubes (Sarstedt AG & Co, Nümbrecht, Germany) and filled up to a total of 10 ml with ultrapure water (Milli-Q, Merck Millipore, Darmstadt, Germany). After microwave digestion of leaf tissues, all samples were fully in solution. Airdried soil samples were sieved (2 mm mesh size). Protocols used for soil extractions were slightly modified from published protocols in order to accommodate high sample numbers and the broad variety of elements to be analysed: The total fraction (nominal) was extracted by mixing a 0.25-g subsample of soil in a mix of 2.25 ml of 37% (w/w) HCl (AnalaR, Merck Ltd, Darmstadt, Germany) and 0.75 ml of 65% (w/w) HNO3, microwave digestion with temperature ramping to 160°C over 15 min and holding for 15 min (Chen & Ma, 2001). Extracts were filtered through filter paper (Whatman No. 1,Brandt,Wertheim,Germany) and adjusted to a total volume of 10 ml with ultrapure water. For the determination of extractable concentrations of elements, 1 g of soil was mixed with 10 ml of 0.1 M HCl in 15-ml round-bottom polypropylene screw-cap tubes using an overhead shaker (150 rpm at RT) for 1 h (Giancoli Barreto et al., 2004;Menzies et al., 2007;Deinlein et al., 2012;Hanikenne et al., 2013). For the measurement of exchangeable concentrations of elements, subsamples of 1 g soil was mixed with 10 ml of 0.01 M BaCl2 in 15-ml round-bottom polypropylene screw-cap tubes using an overhead shaker (150 rpm at RT) overnight (Hendershot & Duquette, 1986;Menzies et al., 2007). These soil extracts were filtered through Whatman No. 1 filter paper, followed by the addition of 1 ml 65% (w/w) HNO3. Element concentrations (Al, B, Ca, Cd, Cr, Cu, Fe, K, Mg, Mn, Ni, P, Pb, S and Zn) were determined in technical triplicates by Inductively Coupled Plasma Atomic Emission Spectrometry (ICP-AES; iCAP 6500 Duo; ThermoFisher, Dreieich, Germany). Every 40 to 50 4 samples, we measured a blank and quality controls (QCs) of an intermediate multi-element calibration standard solution, the appropriate certified reference leaf or soil material (Polish Virginia Tobacco Leaves, INCT-PVTL-6, Institute of Nuclear Chemistry and Technology, Warsaw, Poland; San Joaquin Soil -2709a and Hard Rock Mine Waste -2780, National Institute of Standards and Technology, USA), and either a bulk homogenate of A. halleri leaf tissue, or both a metalliferous and a non-metalliferous soil, as internal laboratory-standardized reference materials (relative standard deviation RSD ≤ 5% among triplicate measurements, among means of independent measurements within a run, and among independent runs, in QCs for all elements). Among measurements of four independent subsamples of leaf material, relative standard deviation for elements present above trace levels was between 1 and 5%. Recoveries that the concentrations of some elements in certified reference materials were far lower than in most of the samples analysed in this study. Data validation and analysis To identify leaf samples potentially contaminated with trace amounts of soil, we generated a dataset of an artificially contaminated (0%, 0.001%, 0.01%, 0.1% and 1% w/w soil) leaf sample (n = 3 replicates) by spiking with extracts from each of 50 soil samples from this survey (25 metalliferous and 25 non-metalliferous soils chosen to reflect the diversity in soil composition) prepared using the leaf digestion method (see above). A Principal Component Analysis (PCA) was conducted on standardized (z-scores) Log10(x + 1) element concentrations determined by ICP-AES to identify elements that contributed the most to principal components, namely Al, Cr, Fe, Ni and Pb, using the function rda from the vegan (V. 2.0-10) R package. Logistic regression models with all five elements, their interactions and all possible element combinations were used to identify leaf samples containing traces of soil using the lrm and validate.lrm functions from the Design (V. 2.3-0) R package (Baxter et al., 2008). To identify samples contaminated with trace amounts of soil, the initial dataset (2,006 leaf samples) was queried, and a total of 34 leaf samples were classified as soil-contaminated and excluded from further analysis. In order to classify sites and individual soil samples as metalliferous or non-metalliferous, element concentrations of individual soil samples or collection sites (median of soil samples) were Log10(x + 1)-transformed, and PCAs were then conducted to identify the elements that consistently showed the largest differentiation in the first dimension irrespective of the soil fraction analysed: Cd, Cu, Pb and Zn (Fig. S2). Their standardized Log10(x + 1) concentrations (z-scores) were used to generate an Euclidean distance matrix using the function vegdist from the vegan (V. 2.0-10) R package, and clustering was performed using Ward's minimum variance with function hclust from the vegan (V. 2.0-10) R package. This procedure was performed on the total, extractable and exchangeable Cd, Cu, Pb and Zn concentrations, and cophenetic correlations and average silhouette widths were calculated (functions cophenetic, using the stats V. 2.15.3 R package, and silhouette from the cluster V. 2.0.1 R package) to decide which fraction (total, exchangeable and extractable soil element concentrations) allowed the most robust assignment of sites to either metalliferous or non-metalliferous character (Fig. S2a, c, e). Accordingly, we obtained 46 metalliferous and 119 non-metalliferous sites based on soil extractable concentrations of Cd, Cu, Pb and Zn (Fig. S2b, d, f; Notes S1), as well as 506 metalliferous and 1,466 non-metalliferous soil samples (Notes S2). To examine whether the 6 timing of sampling may have influenced leaf composition, PCAs were conducted of plant composition (population median concentrations of all analysed elements in leaves were Log10(x + 1)-transformed and standardized (z-scores)) and sampling date (day of year), separately for metalliferous sites and non-metalliferous sites, and jointly for all sites. For none of the principal components (PC) that were statistically significant according to the Kaiser-Guttman criterion did Pearson correlations or scores (loadings) suggest any relationships between sampling date and the concentration of any element in leaves (R 2 or scores ≤ 0.4, Notes S3). Linear regression analyses identified a positive correlation only between Log10(leaf Ca concentrations) and sampling day of the year (all populations, R 2 = 0.019, P < 0.05, n = 165). Histograms displaying probability densities of leaf element concentrations from various datasets were generated using the function multhist (with the plotrix V. 3.5-12 R package). To identify univariate relationships between element concentrations in leaves of individuals and the adjacent soil, linear regressions were generated separately for individuals on metalliferous and non-metalliferous soils using Log10(x + 1)-transformed element concentrations employing the function lm (stats V. 2.15.3 R package). For multivariate relationships between leaf and soil element concentrations, and soil pH, we performed Redundancy Analysis (RDA) models using the function rda from the vegan (V. 2.0-10) R package. Standardized (z-score) Log10(x + 1) soil concentrations (exchangeable, extractable, and total fractions, respectively) and soil pH were used as explanatory variables in models employing Log10(x + 1) leaf concentrations as response variables. The explanatory variables were selected using the step function of the stats (V. 2.15.3) R package, in both forward and backward directions, with significant models with the lowest AIC values being selected and used for final models. We computed adjusted R 2 using the function RsquareAdj from the vegan (V. 2.0-10) R package for the three different soil fractions (Table S1) (Peres-Neto et al., 2006).
To assess the residual within-site variation, we extracted the residuals (using the function residuals from the stats V. 2.15.3 R package) from the RDA between the standardized Log10 (x + 1) soil exchangeable element concentrations, standardized soil pH (both used as explanatory variables, selected as described above), and the leaf element concentrations (as response variables). The residuals were extracted, and 165 PCAs (using the function rda from the vegan V. 2.0-10 R package) were performed on a per site basis. The first three principal components were selected (in sum explaining between 71 and 89% of the total variation) from each of the 165 PCAs, and scores were extracted for all sites per element. Zn and Cd accumulation phenotyping in the growth chamber Seven or more months after transfer to growth facilities, cuttings were made of healthy mother plants by excising small rosettes (5 to 7 leaves), dipping into rooting powder (1% (w/w) indole-3-butyric acid; Rhizopon AA, Rhizopon, Alphen aan den

Supporting Notes S4. Relationships between leaf concentrations of different minerals and
between soil and leaf mineral composition. A Principal Component Analysis (PCA) of leaf mineral composition identified three groups of co-varying elements. The first group consisted of the nutrients P, K, S, Mg, Ca and B, the second group comprised Fe, Al and Cr, and a third group included Cd and Pb (Fig. S4). Furthermore, distinct positions of datapoints from samples taken on metalliferous and non-metalliferous soils suggested a strong influence of local soil type, and consequently inter-element relationships were analysed in the context of soil mineral composition. A more comprehensive assessment of the complex multi-factorial relationships between soil and leaf composition by Redundancy Analysis (RDA) confirmed a strong overall influence of soil exchangeable concentrations particularly of Pb, and also of Cd, Cu, Mn and Zn on their respective concentrations in leaves of A. halleri (Fig. S5a, Table S3, see Fig. 2). We additionally found this for Ni, which can be present as a co-contaminant in the metalliferous soils of our study. With increasing soil pH, leaf Mn concentrations decreased and leaf Ca concentrations increased, thus following soil exchangeable concentrations of these nutrients, as was observed in other species (Marschner, 1995;Kochian et al., 2005). Leaf Al concentrations did not increase with lower soil pH (Kochian et al., 2005), suggesting that A. halleri might effectively restrict the accumulation of Al when it is highly bioavailable. The outcome of this global analysis was dominated by the enormous differences between metalliferous and nonmetalliferous soils. Among metalliferous soils analysed separately, leaf Pb concentrations were positively related to soil exchangeable Pb concentrations, as expected ( Fig. S5b; see Fig. 2c). In addition, leaf Pb and Zn concentrations were inversely related to soil exchangeable Ca concentrations. This suggests a possible competition of Ca, or a possible interference of high soil pH, with leaf accumulation of Zn and Pb, but not of Cd. The dampening of the increase in leaf Zn concentrations with increasing soil Zn concentrations on metalliferous soils (see Fig. 2a) remained unexplained by RDA given the lack of any relationship between exchangeable soil Zn and soil Ca concentrations. Leaf Cd concentrations were correlated with sulphur concentrations in the exchangeable soil fraction. This may reflect a biochemical dependence on the abundance of this macronutrient for the extreme levels leaf Cd hyperaccumulation found on some metalliferous soils. To date, laboratory-based molecular mechanistic studies have not confirmed a biochemical dependence of Zn, Cd or Ni-related extreme traits on sulphur-containing metabolites, in particular phytochelatins (Schat & Kalff, 1992;Krämer et al., 1996;Schat et al., 2002;Meyer et al., 2011), but none of these have addressed differences between accessions from moderately and highly metalliferous soils. Alternatively, soil S availability may positively influence Cd availability for A. halleri. However, this possibility is only hypothetical, because the measured exchangeable soil Cd and S concentrations are largely unrelated with one another, and similarly unrelated are leaf Cd concentrations with exchangeable soil Cd concentrations.
Finally, heightened soil pH was associated not only with lower leaf Mn and Ni concentrations, as was also observed across all soils, but additionally with lowered leaf Cu accumulation. Overall, soil macronutrients, in particular S-versus Mg/P-richness, appeared to be the most important soil constituents influencing leaf composition on metalliferous soils, in conjunction with soil pH.
Among plants on metalliferous soils, we detected an inverse relationship between leaf Ni and Ca, and a few weak positive relationships, most importantly between the concentration of S and Cd, between Pb, Cu and Zn, and between Ca and Fe in leaves. Leaf composition of A. halleri on nonmetalliferous soils showed a dependence on soil pH, and -to a lesser degree -on the concentrations of a group of soil minerals (Mg, Ni, Cr, inversely related with Ca; Fig. S5c).
Heightened soil pH had a moderately negative effect on leaf Zn accumulation, whereas leaf Cd and Pb concentrations did not follow this trend. Taken together, a substantial proportion of the large within-species variation observed on non-metalliferous soils, in particular the variation in leaf Cd concentrations, remained unexplained by soil composition (see Fig. 2; Table S3). In

Fig. S4
Principal Component Analysis (PCA) of leaf element concentrations in A. halleri. Shown are datapoints for 1,972 individuals collected on non-metalliferous soils (n = 1,466 individuals; black) and metalliferous soils (n = 506 individuals; red) (elliptic lines at 95% confidence limits around group centroids). Leaf element compositions constituting PC1 and PC2 loadings are given in bold inside diagram. PCA was conducted employing standardized Log 10 (x + 1) leaf element concentrations (z-scores). , and their matrix correlations with soil exchangeable element concentrations/pH (arrows; characters in italics), for models based on standardized (z-scores) of Log 10 (x + 1) leaf element concentrations, of soil pH and of Log 10 (x + 1) soil exchangeable element concentrations (see Table S3).

Fig. S6
Reproducibility of leaf Zn and Cd accumulation under standardized controlled growth chamber conditions in two independent experiments. Mean leaf (a) Zn and (b) Cd (± SD, n = 5) concentrations in a second independent experiment conducted in Bayreuth are plotted against mean leaf concentrations (± SD, n = 5) of the first experiment conducted in Bochum and shown in Fig. 3 and 4d. Dotted lines correspond to best fits y = 30.8 x 0.545 (R 2 = 0.76) (a) and y = 1.26 x 0.838 (R 2 = 0.77) (b) of means for 25 genotypes (9 originating from metalliferous soils, shown in red; 16 originating from non-metalliferous soiles, black).
Leaf Cd conc.