Functional traits predict species responses to environmental variation in a California grassland annual plant community

Turnover in species composition and community-wide functional traits across environmental gradients is a ubiquitous pattern in ecology, and is generally assumed to reflect shifts in trait optima across these gradients. However, the demographic processes that give rise to these trait turnover patterns at the community level remain unclear. We asked whether shifts in the community-weighted means of three key functional traits across an environmental gradient in a southern California grassland reflect variation in the trait-performance relationship across the landscape. We planted seeds of 17 annual plant species in cleared patches with no competitors, and quantified the lifetime seed production of 1360 individuals. We then asked whether models that included trait-environment interactions help explain interspecific variation in demographic responses to the environment. This allowed us to evaluate whether observed shifts in community-weighted mean traits matched the direction of any trait-environment interactions detected in the plant performance experiment. Our results indicate that commonly-measured plant functional traits help explain variation in species responses to the environment – for example, high-SLA species had a demographic advantage in soils with high soil Ca:Mg levels, while low-SLA species had an advantage in low Ca:Mg soils. We also found that shifts in community-weighted mean traits often reflect the direction of these trait-environment interactions, though not all trait-environment relationships at the community level reflect interactive effects of traits and environment on species performance. Our results support the value of plant functional traits for predicting species responses to environmental variation, and highlight a need for more detailed evaluation of how trait-performance relationships change across environments to improve such predictions.


Introduction
: A) Variation in community-weighted mean (CWM) functional traits across gradients is a common pattern in plant communities, though whether or not such variation in CWM traits reflects shifts in trait optima across environmental gradients. Here we evaluate whether CWM shifts in plant functional traits reflect shifts in trait-performance relationships across key environmental gradients. Panels B-D illustrate how trait-performance relationships might vary across environments. B) The trait-performance relationship may be identical at opposite ends of the environmental gradient, indicating that other factors (e.g. dispersal limitation) might drive observed shifts in CWM traits. We interpret this as a lack of evidence that CWM trait-environment relationships reflect variation in trait optima across the environment. C) The trait-performance relationship may change across the environmental gradient in a direction that is consistent with observed CWM shifts, but the sign of the trait-performance relationship may be the same at either end of the gradient. We interpret this as providing weak evidence that CWM shifts reflect changing trait optima. D) The sign of the trait-performance relationship may change across the gradient, such that species with low trait values have a relative advantage at the low end of the environmental gradient, and vice versa at the high end of the gradient. We interpret this as strong evidence that CWM shifts reflect changing trait optima. optima shift across gradients, but also caution against predicting species responses to environmental 111 variation on the basis of shifts in CWM traits alone.

113
Study system 114 We studied trait-environment relations in the grassland community at the University of California  Quantifying species performance across the landscape 128 In November 2015, before the first major rain storm of the season, we cleared all existing vegetation in 2m 129 X 3m plots at each of our 16 focal sites. At each site, we sowed five replicate plots with the equivalent of 130 20-60 viable seeds each of our 17 focal species (Table S2) on a grid with 15 cm spacing between each 131 species. We collected seeds from hundreds of plants growing across Sedgwick Reserve in the spring prior 132 to this study, and seeds were homogenized among sources before planting to ensure that local adaptation 133 (Rajakaruna and Bohm 1999) or maternal effects (Germain and Gilbert 2014) did not systematically drive 134 variation in plant performance across sites in our experiment.

135
In February 2016, we counted the number of germinants of each focal species in our experimental plots, 136 and thinned each plot to leave only two individuals of each focal species. In March, we further thinned 137 each point down to a single individual of each species, and weeded around this focal individual to ensure 138 that it was not competing with other plants in a 15cm radius. Between April-June 2016, we quantified the 139 total seed output of each focal individual in our experiment, for a total of 1360 individuals (17 species * 16 140 sites * 5 plots per site) tracked across the environment (see Appendix S1 for details on how we estimated 141 total seed output). This design let us quantify the germination rate and the per-germinant seed production 142 6 (fecundity) in the absence of competitors for each species at each site. Both of these vital rates are known to be important determinant of annual plant demography in this community (Levine and HilleRisLambers 144   2009), but we focus only on fecundity as a measure of species performance in the remainder of this study 145 because the functional traits we measured most clearly relate to the growth of plants after germination.

146
Measuring compositional turnover across the landscape 147 In Spring 2017 (the year after the experimental assessment of plant performance), we surveyed five 148 undisturbed plots (0.5x0.5m) adjacent to the experimental plots at each of our 16 sites to characterize the 149 vascular plant community composition. These community survey plots were spaced evenly on a 10m 150 transect located alongside the cleared plots in which we had experimentally quantified plant performance 151 in the prior year. In each plot, we visually estimated the total (absolute) cover of each of species in early 152 April, and again in early June.  Hordeum murinum, Micropus californica, and Vulpia microstachys. We followed the protocols detailed in Kraft 160 et al. (2015) to measure traits from 5-8 individuals growing in 0.7*0.7m plots at three of the matrix sites in 161 our experiment, which we had sowed with seeds of all 17 annual plant species for a total sowing density 162 of 8g viable seeds/m 2 . In Spring 2017, we measured the same set of functional traits on the 38 of the most 163 common annual plant species encountered within the community composition plots (of the species for 164 which we could not measure at least one of the focal traits, mean cover of these species in sites where they 165 were present was < 5%).

166
Our analysis focuses on three traits that capture distinct dimensions of plant ecological strategies and 167 that were largely uncorrelated in a principal components analysis of the traits we measured (Fig. S1): Environmental sampling 181 We quantified various soil chemical and physical characteristics to identify the primary axes of 182 environmental variation among our study sites. We measured gravimetric water content ((weight of fresh 183 soil -weight of dry soil)/weight of dry soil) in the early-and mid-growing season (March and April, 184 respectively), and summarized across these measurements to estimate the average soil moisture at each 185 site. At each site, we also collected soil for analysis by A&L Western Agricultural Laboratories (Modesto, 186 CA) for a variety of soil chemical and physical properties: soil organic matter, P (Weak Bray and Olsen 187 methods), K (ppm), Mg (ppm), Ca (ppm), Na(ppm), pH, CEC, NO 3 , SO 4 , NH 4 , and soil texture (sand, silt, 188 and clay content). We collected the soil for these analysis from three points arranged in between the five 189 experimental plots, and homogenized within site prior to analysis. We also programmed iButtons (Maxim 190 Integrated) to log temperature at 2-hr intervals, and used these data to quantify the average daily Quantifying community-weighted trait turnover across the landscape 197 We used the community composition and functional trait measurements to calculate the 198 community-weighted mean (CWM) trait values, which represent the mean trait value of all species 199 growing at a site, weighted by the species' relative cover. We calculated the CWM for each trait (t) at each 200 of our 16 sites (s) by averaging across the CWM of the five plots p at each site as follows: model. We considered trait-environment interactions that were significant in the model, but whose slope 223 did not change sign across the environmental gradient, as weak evidence that CWM trait shifts reflect 224 shifts in trait optima across the landscape (Fig. 1C). If the sign of the trait-performance sign shifted in the 225 direction predicted by CWM trait shifts, we considered this as strong evidence that CWM trait shifts 226 reflect shifts in trait optima across the landscape (Fig. 1D).

227
We conducted all analyses in R v. 3.6.3 (R Core Team 2020) and provide code to recreate all analyses in 228 appendix S2. All data are available as supplementary files and will be deposited in to an archival 229 repository prior to publication.

241
The fixed effects of our GLMM with all the main and interactive effects explained 18% of the variation in 243 seed production (Marginal R 2 = 0.18), and the random effects of species and site explained an additional 244 19% of the variation (Conditional R 2 = 0.37). The model included significant positive main effects of soil 245 Ca:Mg (p = 0.016) and soil depth (p = 0.016), indicating that seed production was higher in sites with 246 higher Ca:Mg and deeper soils, irrespective of plant traits (Fig 2). The main effect of maximum height was 247 also significant and positive (p = 0.005), indicating higher seed output from larger-statured plants across 248 the environmental gradient (Fig 2).

249
The model also provided evidence that three out of the four significant relationships between CWM which may give low-SLA species a relative advantage in low-Ca:Mg sites, which also tended to have 299 lower soil moisture in this system (Fig. S3).

300
We also found a negative relationship between CWM SRL and sand content (Fig. 3C). This relationship Arizona. This discrepancy may have arisen in part because in our system, soil sand content was generally 304 13 much higher on serpentine hummocks that were also characterized by low soil moisture and organic 305 matter (Fig. S3). In this context, the negative relationship between CWM SRL and sand content is 306 consistent with the more general expectation of low SRL indicating a resource-conservative strategy that 307 allows plants outperform species with resource-acquisitive strategies in more stressful conditions (Reich 308 2014). Moreover, our analysis of trait and environmental predictors of seed production provides strong 309 evidence that this community-level pattern is in part driven by low-SLA species having higher intrinsic 310 fecundity in sandy soils, and vice-versa in soils with low sand content (Fig. 3C). Understanding the in various demographic processes will be critical for understanding how plant traits determine overall 335 population growth rates and this influence community assembly processes across landscapes.

336
Although three three of the four CWM trait-environment correlations in this study seem to at least 337 qualitatively reflect the direction of the trait-environment interaction in terms of species' intrinsic 338 fecundity, we did not find such evidence for the positive CWM SLA-soil depth correlation (Fig. 3B,F). This Our analysis also allows us to ask whether any trait-environment interactions mediate variation in 350 species performance but do not appear to turn over across the environmental gradient at the community 351 level. We did not find any evidence for trait-environment interactions influencing species performance 352 that did not manifest in CWM trait turnover (Fig. S4). This suggests that in our annual grassland system, that drive trait turnover patterns at the community level.

376
A second limitation of our study is that we were unable to account for the possibility that intra-specific 377 trait variation (ITV) driven by local adaptation, phenotypic plasticity, or maternal effects -processes that Ca:Mg (Fig. 3E) suggests that ITV may be structured such that individuals of the same species growing in 384 soils with higher Ca:Mg build higher-SLA leaves than conspecific individuals on low-Ca:Mg soils.

385
Understanding how the spatial structure of ITV differs between species may be critical for predicting Max height x Soil depth* Figure S5: 3D interaction surfaces for all nine trait-environment interactions in our GLMM of seed production as a function of trait and environment predictors. Plots labeled with an asterisk indicate significant interaction terms. 24