Soil nutrients and precipitation are major drivers of global patterns of grass leaf silicification

Abstract Grasses accumulate high concentrations of silicon (Si) in their tissues, with potential benefits including herbivore defense, improved water balance, and reduced leaf construction costs. Although Si is one of the most widely varying leaf constituents among individuals, species, and ecosystems, the environmental forces driving this variation remain elusive and understudied. To understand relationships between environmental factors and grass Si accumulation better, we analyzed foliar chemistry of grasses from 17 globally distributed sites where nutrient inputs and grazing were manipulated. These sites span natural gradients in temperature, precipitation, and underlying soil properties, which allowed us to assess the relative importance of soil moisture and nutrients across variation in climate. Foliar Si concentration did not respond to large mammalian grazer exclusion, but significant variation in herbivore abundance among sites may have precluded the observation of defoliation effects at these sites. However, nutrient addition consistently reduced leaf Si, especially at sites with low soil nitrogen prior to nutrient addition. Additionally, a leaf‐level trade‐off between Si and carbon (C) existed that was stronger at arid sites than mesic sites. Our results suggest soil nutrient limitation favors investment in Si over C‐based leaf construction, and that fixing C is especially costly relative to assimilating Si when water is limiting. Our results demonstrate the importance of soil nutrients and precipitation as key drivers of global grass silicification patterns.

conditions, the organic plant tissue matrix is dissolved by NaOH and undergoes oxidation by the peroxide; organic material remains bound to the oxidant, and Si is then released from the matrix and dissolved by NaOH (Elliott & Snyder 1991). Immediately following digestion (< 24hr), 1 ml of 5 mmol NH4F (Kraska & Breitenbeck 2010) was added to each sample and they were diluted to 50 ml in distilled-deionized water. The fluoride ion added in this step contributes to the complete dissolution of Si compounds, which may not be totally dissolved during the previous strong-base digestion. The resulting solution is then further diluted prior to Si quantification by inductively couple plasma optical emission spectroscopy (ICP OES). In this case, 100 µl of solution is diluted to 10 ml with 1% v/v HCl. Recovery following digestion was validated by use of a certified reference material (Community Bureau of Reference, BCR, Reference material No.

129, hay powder).
A Prodigy ICP OES system (Teledyne Leeman Labs, Hudson, NH, USA) composed of an automatic sampler, a double-pass spray chamber and a concentric nebulizer was used to quantify Si content of selected calibration samples. The operating conditions were: axial view mode, radio-frequency applied power 1.3 kW, plasma gas flow rate 18 L/min, nebulizer pressure 30 psi, were used to build the analytical calibration curve. The limit of detection (LOD) for the ICP OES determination, calculated according to IUPAC's recommendations as 3 times the standard deviation of the blank solution (n = 10) divided by the calibration curve slope, was 0.05% silicon in dry mass. Samples which fell below the LOD were eliminated from our calibration dataset.

NIRS Model
We fit our NIRS calibration model using a Partial Least Squares (PLS) regression relating leaf Si to spectral variation, using the "plantspec" R package (Griffith & Anderson 2018).
Representative calibration (n = 345) and validation (n = 87) data subsets were selected for wet chemistry (ICP OES) using a modified Kennard-Stone algorithm termed "SPXY" which incorporates variation in both the spectra and the Si values (Snee 1977, Saptoro et al. 2012. During model development, we considered whether model fit improved when focusing on different continents, plant functional types, or when using a variety of spectral preprocessing steps and specific spectral regions. The model with the lowest Root Mean Squared Error of Prediction (RMSEP = 0.52; 25 latent vectors) was one in which spectra were first preprocessed with Vector Normalization and then restricted to between 7500 to 6100 cm -1 and 5450 to 4600 cm -1 which corroborate the wavenumbers used by Smis et al., (2014). In addition, the model was greatly improved by the removal of bryophyte and Mt. Caroline (Australia) samples, which each had unique spectral properties. Therefore, when analyzing the final dataset, neither bryophytes nor predicted data from Mt. Caroline were included. The final calibration model was validated on the test set and performed well (validation R 2 = 0.83; calibration R 2 = 0.86) (Supplementary Figure S1). Tables   Table S1. Sites included in data analysis. Site locations are displayed in Fig. 1 (main text)  . An observed slope more negative than the expected distribution of slopes indicates that a particular element is being exchanged for carbon. Figure S3. Species-specific Si-accumulation of grass species from the site Chichaqua Bottoms.

Supplementary
Andropogon gerardii is a C4 grass, and the remaining 5 species are all C3.