Linking grassland plant diversity to species pools, sorting and plant traits

Summary

1. Variation in species pools can affect plant diversity, but it remains unclear whether the magnitude of the response varies because of resource availability, community invasibility or other environmental factors, and whether colonization along environmental gradients reflects niche-based species sorting or neutral processes.

2. We hypothesized that unimodal diversity–productivity patterns in grasslands are dependent on species pools, with peak richness occurring at intermediate productivity due to species sorting associated with species traits. We used a seed-addition experiment to test the influence of immigration on plant species richness across multiple grasslands (old fields), each of which encompassed a broad range in productivity. We then tested whether species sorting occurs during colonization, if this varies with site productivity, and whether sorting patterns are associated with species traits 1 and 4 years after seed addition.

3. Augmentation of species pools increased species richness across fields with the greatest increase in colonization in all fields at sites with intermediate productivity. The increased diversity following species pool augmentation during colonization (year 1) was associated with seed characteristics, seedling growth rates and allocation, and water use, suggesting that colonization and seedling establishment were the result of species sorting.

4. Although there were some changes in trait–environmental relationships between the first and fourth growing season, patterns that were observed in the first year were still apparent in the fourth year, highlighting the persistent importance of species sorting and traits that affect colonization on species distribution and diversity patterns in grasslands.

5. Synthesis. Our experimental results suggest that species richness and richness–productivity relationships in successional grasslands are at least partially determined by species sorting and trait–environmental relationships occurring during colonization.

Introduction

Explaining why species diversity varies within and among plant communities remains a formidable unresolved question in ecology. The limited success of many species interaction models in explaining species diversity patterns has led to investigation of the potential constraints imposed on local diversity by the species pool in the surrounding area (Ricklefs 1987; Cornell & Lawton 1992; Leibold et al. 2004). Species pools can influence local species interactions by determining the number and identity of species available to interact and by rescuing species from extinction through immigration (van der Putten et al. 2009). Mounting evidence suggests that the size and composition of the regional species pool influence plant diversity (Tilman 1997; Zobel 1997; cf. Foster 2001). These species pool effects appear to interact with local abiotic and biotic factors to control observed diversity patterns in many systems, but particularly in grasslands (Zobel et al. 2000; Foster 2001; Houseman & Gross 2006). The growing acknowledgement of the importance of local species interactions and immigration has led to the development of metacommunity theory that specifically addresses how the outcome of local species interactions is modified by migration between communities and shows how local and regional scale processes influence diversity (Leibold et al. 2004).

An important element of local–regional or metacommunity models of diversity is identifying conditions under which species pools have strong or weak effects on diversity. The results of seed-addition experiments in grasslands have shown that the influence of the species pool can vary widely (Turnbull, Crawley & Rees 2000). Foster et al. (2004) proposed that underlying site productivity mediates the importance of the species pool for local diversity, with stronger species pool effects at low to intermediate productivity. Although evidence for this view is growing (Foster et al. 2004; Stevens et al. 2004; Houseman & Gross 2006), there remain substantial differences among these studies. Some studies have found that colonization from augmented species pools is lower in high than intermediate productivity sites (Foster & Dickson 2004; Houseman & Gross 2006), but that sites with low productivity show inconsistent responses (Foster et al. 2004; Houseman & Gross 2006). It is unclear whether this incongruence reflects site-specific differences among studies such as site history, the magnitude of environmental variation found along the resource gradient or the invasibility of the established plant community.

Moreover, while augmenting the species pools has been shown to increase local diversity, it is not clear whether the observed increase in species richness is accounted for by a random subset of the species pool or reflects species sorting patterns driven by trait–environment relationships (Lavorel et al. 1997). Differing effects of species augmentation on local richness across sites that vary in productivity (or some other environmental characteristic) may reflect species sorting based on variation in traits among species in the species pool. If specific traits confer different establishment probabilities depending on environmental conditions, then we would expect to see changes in both the richness and composition of species that colonize sites. In contrast, colonization success may be dependent on stochastic variation during immigration resulting in a random selection from the species pool (Hubbell 2001). Neutral processes could lead to variation in diversity along resource gradients if seedling emergence or mortality rates for all species varied with productivity. Nutrient or moisture limitation at low productivity, and light limitation at high productivity, could decrease emergence or increase seedling mortality relative to intermediate productivity sites. The higher recruitment at intermediate productivity could reduce the number of rare species lost to stochastic extinction leading to peak richness at intermediate productivity. Thus, a unimodal productivity–diversity pattern could occur without any relationship between species identity (or traits) and productivity. Despite the growing interest in local–regional or metacommunity concepts as drivers of diversity, there is little experimental evidence as to whether neutral or niche-based colonization processes determine the observed diversity patterns across productivity gradients.

In this study, we test two hypotheses as to what determines productivity–diversity patterns in grasslands: that (i) the strength and shape of these patterns are dependent on the species pool and (ii) such patterns reflect species sorting associated with species traits. A test of the first hypothesis requires experiments across multiple sites with similar environmental gradients using the same species pool to determine if there are generalities in the strength of the local–regional diversity interaction related to productivity. We report here the results of such a cross-site experiment designed to examine this relationship in four successional grasslands that had substantial within-site variation in productivity and species composition and therefore to investigate the generality of the colonization–productivity relationship (e.g. the unimodal pattern; Houseman & Gross 2006). Second, we tested whether the species that colonized these sites represented a random selection from the species pool or species sorting in relation to productivity. Evidence for species sorting along productivity gradients has primarily come from observational studies (Whittaker 1967; Weiher, Clarke & Keddy 1998; Ackerly & Cornwell 2007), but it is unclear whether the distributional patterns reflect trait–environment sorting patterns or are a consequence of differences in dispersal limitation along the environmental gradient. Finally, we tested whether any of the observed species sorting patterns in our field experiment were related to species traits, which were independently measured in a greenhouse study. Although these experiments cannot reveal the extent to which metacommunities operate in this system, they provide a critical test of whether colonization processes – which are an inherent part of metacommunity dynamics – reflect niche-based or neutral processes.

Materials and methods

Field experiments

We established seed-addition experiments in four successional grasslands (old fields) located within 7 km of Michigan State University’s W.K. Kellogg Biological Station in SW Michigan. All four fields had been abandoned from agriculture 10–40 years prior to our experiment and all were dominated by a mixture of native and non-native herbaceous perennials typical of mid-successional fields in the Upper Midwestern USA (see Appendix S1 in Supporting Information). Within each field, there was a pronounced productivity-topographic gradient (200–700 g m⁻² in above-ground biomass along a slope of<200 m) that corresponded to differences in soil fertility and moisture. Species composition also varied across the gradient and among sites (Houseman 2004). Based on topographic position along the gradient, we selected three sites in each field that differed in productivity (low, medium/intermediate, high) and species composition (see Appendix S1).

In each of the three productivity levels (sites) in each field, we established a grid of sixteen 1.2 × 1.2 m plots separated by 0.5–1 m buffer strips. We augmented species pools by adding seeds of 40 species to eight randomly selected plots in each site. The species added were native to Michigan and encompassed a range of plant traits (see Appendix S2). All but four species were initially absent from all fields in the study. We sowed 1 g m⁻² of each species into the seed-addition plots in April 2003. We used equal seed mass rather than number to account for differences in potential immigration rates that are likely to be related to seed mass. When we added seeds, we agitated the litter – but not the soil – of all plots with a stiff rake to assure that the introduced seeds reached the soil surface. Because of a spring drought, all plots were watered once in late April – an amount equivalent to a single rainfall event of c. 2 mm; no additional water was added to the plots thereafter.

We determined species composition in all plots after one (late August–early September in 2003, year 1) and four (September 2006, year 4) growing seasons following the seed addition to determine if initial patterns of colonization were maintained. In both years, we determined species presence and per cent cover in the central 1 m² of each plot. Per cent cover was estimated by using reference cards of known area. Mean species richness in seed-augmented plots was compared with that in control plots to which no seeds were added. In year 4, we were only able to sample three of the four fields as one of the fields was ploughed and replanted by the landowner.

To relate diversity responses to environmental conditions, we measured light availability at ground level in four ancillary plots adjacent to the three experimental plots in each field to provide a metric of productivity differences within and among fields. Light availability at ground level is a reasonable surrogate for productivity in these sites (Houseman 2004) and is also a direct measure of a resource thought to be important for plant competitive interactions along productivity gradients (Grace 1999). We measured light interception by the plant canopies in late August and early September with a Decagon Sunfleck Ceptometer (SF-80) (Pullman, WA, USA) in the central 0.5 × 1 m of ancillary plots. Ceptometer readings (photosynthetic photon flux density, PPFD) were taken below the plant canopy (ground level, but above the litter) every 10 cm along one edge of the plot and averaged to provide an estimate of light availability for each site. All readings were taken within 2 h of solar noon and were expressed as a percentage of PPFD measured above the plant canopy.

Assay of species traits

To relate species traits to colonization and establishment, we used greenhouse studies to determine a suite of morphological and physiological traits thought to be important determinants of colonization (Weiher et al. 1999; Cornelissen et al. 2003) for all of the species added in the field experiments. Twelve of the 40 species either failed to germinate or grow in the greenhouse trials, so our trait analysis is restricted to the remaining 28 species. The absence of some species that germinated in the field but failed to grow in the greenhouse could bias the detection of trait–environment relationships particularly if they belong to certain functional groups. However, only two species (Baptisia leucantha and Oenothera biennis) that had any indication of significant sorting in the field (see results) failed to grow in the greenhouse. The former species is a perennial legume whereas the latter is a biennial, non-leguminous forb. Consequently, the omission of these two species should not alter the detection of productivity–diversity patterns.

We estimated germination rates and viability (per cent germination) in an environmental chamber set at 25 °C on a 12/12 light/dark cycle. Seeds of each species were placed in covered petri plates filled with wet, sterilized sand and assayed daily for germination for 63 days. To assay growth characteristics, germinated seeds were transferred to plug trays (each cell 1.4 × 1.4 cm) filled with field soil. Following seedling establishment in the plug trays, five seedlings of each species were transplanted to individual pots (3.8 cm width × 21 cm depth; 164 cm³) filled with soil (90% sand, 3.6% silt, 6.4% clay) collected from a nearby field. An additional five seedlings were harvested and dried to determine the initial biomass for growth rate calculations (see below).

The transplanted seedlings were kept well-watered and fertilized three times per week with ammonium nitrate. Following transplanting, each seedling received an initial dose of N (ammonium nitrate) dissolved in water and equivalent to 1.1 g m⁻². Thereafter, N was added at this same level three times per week for a total of 38 g m⁻² applied over the entire experiment. To reduce variation in air and soil temperature on seedling performance in these small pots, the pots were placed in insulated boxes filled with vermiculite that was periodically watered, which both insulated the pots and reduced desiccation. As natural sunlight decreased in the fall, we supplemented light with high-pressure sodium lights on a 12/12 cycle. After 14 weeks, plants were separated into leaf, root and shoot material, and dried for 48 h at 60 °C. Dry mass was used to quantify SLA (specific leaf area) and root : shoot allocation. Seedling growth rates were determined by comparing the final to initial seedling biomass using the formula: GR = ln(final biomass − initial biomass)/time. To determine species impacts on soil resources (a surrogate of R*), we measured soil inorganic N and moisture in the pots at the time of harvest following methods described in Robertson et al. (1999). Water and N use were determined from the amount remaining in the soil at harvest.

Statistical analysis

To test whether productivity influences species pool–richness relationships across fields, we first compared the response to seed addition across the three productivity levels (sites) in each of the four fields with a three-way mixed-model anova (sas v9.0; SAS Institute Inc., Cary, NC, USA). Relationships between species richness and light availability (i.e. a surrogate for above-ground biomass or productivity) were tested with regression (systat; SYSTAT Software Inc., Chicago, IL, USA). We used the Mitchell-Olds and Shaw (MOS) test to determine whether these nonlinear relationships had a clear ‘peak’ within the data range (Mitchell-Olds & Shaw 1987). To examine the general richness–productivity relationships across environments, we combined data across the three sites from the four (year 1) or three (year 4) fields. We used regression to test for a relationship between light availability and species richness and tested for both linear and quadratic relationships in the control (resident species) and augmented species pool (resident and colonized species) treatments.

We used Indicator Species Analysis (ISA; Dufrene & Legendre 1997; McCune & Grace 2002) to test whether species differentially colonized sites in each field that differed in productivity. Because we were looking for a link between species sorting and a unimodal productivity–diversity relationship, we examined how each site contributed to the unimodal productivity–diversity patterns rather than relying on the original classification of sites (low, medium or high productivity). This was performed by sequentially removing sites from each end of the gradient until the light–richness relationship was no longer a significant unimodal relationship. Those sites required to maintain the unimodal patterns were classified as either low (high light) or high (low light) productivity. Using this approach led to a reclassification of the low productivity site in field 3 to ‘medium’ productivity using the full array of sites. We then performed an ISA using these three site types (low, medium and high productivity) based on presence–absence data – as we were most interested in how species sorting related to richness–productivity relationships. Significant association between each species and a particular site type was tested with a permutation test (1000 runs) based on the 24 species that were observed in at least one plot in the second sampling (2006), 4 years after the seed augmentation. Focusing only on the species that survived to the fourth year reduced the impact of any ‘spurious relationships’ and provided a stronger test of the trait–environment relationships.

We then used a fourth-corner analysis (FCA) to test whether species traits measured in the greenhouse were related to species colonization patterns observed in the field (Legendre, Galzin & HarmelinVivien 1997). We focused only on those species that we identified as sorting among environments as we were interested in whether species sorting patterns were related to species traits. We included Lespedeza capitata with the ‘sorting species’ as it was extremely close to the traditional cut-off value (ISA, P = 0.057) and this allowed the inclusion of three legumes in subsequent trait analysis. Because species traits are inherently inter-correlated, we reduced the number of trait variables with PCA after standardizing all variables (McCune & Grace 2002). Although this multivariate approach can create complexity in the interpretation of the axes, it has the advantage of allowing us to detect fundamental trait constellations in which multiple traits operate together to determine functional attributes while avoiding multiple tests of potentially inter-correlated traits.

To perform the FCA, we calculated three matrices. The first matrix was a species × plots matrix based on presence–absence in the field plots after 1 and 4 years. The second matrix was a plots × environment matrix using light availability to characterize the environmental gradient. The final matrix was a species × trait matrix using the first four PCA axis scores for species traits. From these three matrices, a fourth matrix was derived where species traits were related directly to environmental conditions. Significant association was tested with permutation tests (10 000 runs). All ordinations were performed with pc-ord v4.0 (MjM Software, Glenden Beach, OR, USA).

Results

Species pools and richness

Seed addition increased species richness in all four fields (F = 320, P < 0.0001; Fig. 1) in year 1. Because there was a three-way interaction between field, site (low, intermediate and high productivity) and seed addition (F = 6.31, P < 0.0001), we used contrasts to compare the effect of seed addition in each of the four fields. In each field, there was a significant effect of seed addition on species richness (F > 33, P < 0.0001 in all fields). However, there was no effect of seed addition on species richness in the high productivity sites (<15% of full light) in two of the fields (Fig. 1, fields 2–3). Although the shape of the richness–productivity curves differed among sites for the control plots, all four fields showed similar unimodal patterns with a peak at intermediate productivity following seed addition (light availability; MOS test P < 0.05 for each field).

Figure 1.  Species richness as a function of light availability at ground level (a surrogate for productivity) in four grasslands plotted either (a) separately in year 4 or (b) collectively in years 1 and 4 for control (resident species, X, dashed lines) and seed pool augmented plots (resident + added species: circles, solid lines) plots. Points are plotted with a random jitter for clarity. Lines are best fits of either linear or quadratic models. The difference in species richness between control and augmented species pools was always significant (P < 0.05) except for the low end of the light gradient (high productivity sites) in fields 2 and 3. Note that values for % Light availability are plotted from high to low corresponding to increasing productivity.

When data from all fields were combined (Fig. 1), species richness was unimodally related to light availability in year 1 in both the control (resident species; model: R² = 0.42, d.f. = 2, 93, P < 0.001; quadratic term t = −4.17, P < 0.001; MOS P < 0.05) and seed-added plots (model: R² = 0.59, d.f. = 2, 93, P < 0.001; quadratic term t = −5.85, P < 0.001; MOS P < 0.001). However, seed addition increased the magnitude of the unimodal pattern. Four years after the seed addition (2006), there was still a stronger unimodal relationship between species richness and productivity in the seed-added plots (model: R² = 0.36, d.f. = 2, 69, P < 0.001; quadratic term t = −4.87, P < 0.001; MOS P < 0.001).

Species sorting

To test whether species sorting or neutral processes contributed to the enhanced unimodal productivity–diversity pattern in these fields, we used an ISA with species allocated to low, intermediate and high productivity sites as defined above. Plots with >60% light were classified as low productivity, whereas those with <12% light were designated as high productivity and the remaining plots as intermediate productivity plots (see Materials and methods). We found that 15 of the 40 added species had an affinity (P < 0.06, Appendix S3; Fig. 2) for a particular productivity site type by the fourth year indicating significant sorting patterns (χ² = 10, d.f. = 2, P < 0.05).

Figure 2.  Results from the Indicator Species Analysis (ISA) based on presence–absence data for 15 species that showed significant sorting in relation to productivity 4 years after seed addition (Appendix S3). Data are aggregated over all three fields. Species codes are as follows: KUHEUP = Kuhnia eupatorioides, SORNUT = Sorghastrum nutans, SCHSCO = Schizachyrium scoparium, ANDGER = Andropogon gerardii, PANVIR = Panicum virgatum, HELHEL = Heliopsis helianthoides, ASCTUB = Asclepias tuberosa, MONFIS = Monarda fistulosa, RUDHIR = Rudbeckia hirta, RATPIN = Ratibida pinnata, LESCAP = Lespedeza capitata, ECHPUR = Echinacea purpurea, BAPLEU = Baptisia lactea, OENBIE = Oenothera biennis, DESSPP = Desmodium spp.

The sorting patterns found after 4 years reinforced those that were apparent after one growing season. After a single growing season, 8 of the 27 species [Andropogon gerardii Vitman, Amorpha canescens Pursh, Asclepias syriaca L., Asclepias tuberosa L., Desmodium spp., Echinacea purpurea (L.) Moench, Lespedeza capitata Michx. and Monarda fistulosa L.] showed significant sorting patterns between low and intermediate productivity sites. One species was more common in low productivity sites and seven in intermediate productivity sites (Appendix S3, Fig. 2). Between years 1 and 4, 10 additional species [Baptisia lactea (Raf.) Thieret, Echinacea purpurea (L) Moench, Heliopsis helianthoides (L.) Sweet, Kuhnia eupatorioides L., Oenothera biennis L., Panicum virgatum L., Ratibida pinnata (Vent.) Barnhart, Rudbeckia hirta L., Schizachyrium scoparium (Michx.) Nash and Sorghastrum nutans (L.) Nash] developed species sorting patterns. Three species (Amorpha canescens Pursh, Asclepias syriaca L. and Helianthus maximiliani Schrad.) that initially exhibited sorting pattern (year 1) did not have a differentiated distribution in relation to productivity in the fourth growing season.

Species traits

To examine whether there were identifiable species traits related to the observed differences in species sorting patterns across the productivity gradient, we related traits – assayed in laboratory and greenhouse studies and summarized by PCA axes – to initial establishment (year 1) and persistence (year 4) patterns in the field experiment by a FCA (see Materials and methods). The first three axes of the PCA for species that established in the first year explained 76% of the variation in species traits. The FCA indicated positive relationships between each of the PCA axes and productivity in the first year (Table 1). Based on the loading of species traits on the PCA axes, species with high growth rates and N-use (PC1) were more successful in colonizing higher productivity sites. There was a weak trend for species with larger seed masses, higher germination rates and higher root : shoot allocation (PC2) and decreased germination rate (PC3) to be more successful as productivity increased.

Table 1.   Results of the PCA for the species identified as showing significant sorting in response to productivity in the Indicator Species Analysis after the first and fourth growing season

	Year 1			Year 4
	PC1	PC2	PC3	PC1	PC2	PC3
The upper portion of the table contains the Pearson correlation between the PC axis and productivity (as measured by light penetration at ground level) based on the fourth-corner analysis (FCA). The P-value (bold) is calculated by permutation (Legendre, Galzin & HarmelinVivien 1997). The lower portions of the table contain the loading of species traits on the PC axis and the percentage of variation explained by each axis. SLA, specific leaf area.
FCA
R	0.28	0.07	0.07	0.12	0.01	0.28
P-value	0.0001	0.0416	0.0343	0.0002	0.3814	0.0001
Shoot growth rate	−0.461	0.07	−0.028	−0.452	0.042	0.29
Root growth rate	−0.208	−0.444	0.492	−0.111	0.639	0.132
Root : shoot	0.277	−0.524	−0.117	0.197	0.548	−0.264
Loading of traits on PCA axes
SLA	0.089	0.125	0.282	0.135	−0.102	−0.274
H2O available	0.306	−0.09	0.239	0.044	0.327	0.475
NO3 available	0.462	0.005	−0.284	0.488	−0.135	−0.139
NH4 available	0.461	−0.034	−0.219	0.49	0.017	0.141
Seed mass	0.125	−0.491	0.336	0.291	0.354	−0.177
% Germination	−0.218	−0.383	−0.398	−0.328	0.163	−0.432
Germination rate	0.278	0.333	0.454	0.234	−0.045	0.52
R^2 for PC axes	39.6	21.6	14.7	33.9	22.2	20.6

Some, but not all, of the traits we identified as important to initial establishment were associated with persistence through the fourth year. Species with higher growth rates and Nitrogen use were found at higher productivity sites (PC1; Table 1). Likewise, species with increased water use and total germination were associated with higher productivity sites (PC3). However, PC2 – which was associated with root growth rate and root : shoot allocation – was not significantly related to establishment patterns in the field.

Discussion

Our results show that species richness was limited by immigration in all four successional grasslands and the magnitude of the response to species addition varied within all four fields with productivity. We observed the largest increases in species richness in sites of intermediate productivity in all fields. These productivity-dependent differences in species richness were observed in the year following seed addition and persisted for 4 years. The higher recruitment in sites of intermediate productivity is partially explained by species sorting that can be linked to variation in species traits, suggesting an important role for niche processes in determining patterns of species diversity in these grasslands.

Tests of immigration and richness

The consistent response to seed augmentation that we observed across these four sites (Fig. 1) has important consequences for our understanding of the role of colonization and regional source pools in determining diversity patterns in grasslands. Our results support those of other studies conducted in single sites, which found that immigration varies with resource conditions; however, the magnitude and shape of the species pool–richness relationship varies among these studies (Foster et al. 2004; Stevens et al. 2004; Houseman & Gross 2006). One explanation for the differences among studies is that site history and resident species composition influence the species pool–diversity relationship. However, among our four fields, there were striking differences in species composition, yet peak richness consistently occurred in each field where there was c. 40% light availability. This consistency among sites suggests that above-ground productivity, more than species composition, is important in determining the shape of the productivity–diversity pattern and that this relationship will change in response to increases in the regional species pool (Huston 1999; Houseman & Gross 2006; Partel & Zobel 2007). This is not to say that site history is not important in determining diversity and composition – particularly in successional communities (Fargione, Brown & Tilman 2003; Houseman et al. 2008). Among the four grasslands we studied, there were differences in the magnitude of the response to seed augmentation, particularly in high productivity sites (see Fig. 1) suggesting that high productivity sites are more likely to show different community states than low to intermediate productivity sites (Houseman et al. 2008).

The consistent increase in species richness and the maintenance of the relationship with productivity in these fields with or without species pool augmentation suggest that environmental conditions are strong constraints on richness (Fig. 1). The reduced colonization at low and high productivity sites in these grasslands is likely due to strong abiotic and biotic filtering along the gradient (Belyea & Lancaster 1999). In an earlier study, we estimated the magnitude of abiotic and biotic filtering in relation to productivity gradients in old fields (Houseman & Gross 2006). In this previous study – conducted adjacent to plots in field 2, we showed that abiotic filters likely reduced colonization at low productivity, as many species were unable to establish in the absence of competitors but differences in colonization rates in plots with and without competitors were small. In contrast, at high productivity, many species capable of establishing in the absence of competitors were unable to establish in the presence of competitors, suggesting strong biotic filtering of species. Clearly, environmental variation remains an important factor in determining local diversity even when immigration rates are high, or experimentally augmented.

Tests of species sorting and plant traits

Determining whether the unimodal pattern of species richness and productivity under higher rates of immigration are related to species traits varying among environments or to neutral processes requires consideration of how species performance varies in response to environmental gradients. A unimodal pattern – and higher colonization at intermediate productivity – can occur if mortality rates for all species are lower at sites of intermediate productivity and higher at low and high productivity sites. If that were true, then species richness is not dependent on species traits but rather neutral processes that determine mortality across environments. Alternatively, if species performance (colonization or growth) varies across environments and fewer species are capable of colonizing or persisting under the environmental conditions found, then this is evidence for species sorting. If there is a match (or some concordance) in the traits of species and the environments they can colonize, then this is further evidence that niche-based processes are important in determining the productivity–diversity patterns we observe.

We not only found evidence for species sorting in the first year (colonization), but also these effects persisted for four growing seasons. The ISA indicated that 15 of the 27 species that established in this experiment exhibited differential patterns related to productivity; five species preferentially colonized/established in sites with low productivity and 10 species were observed more often in intermediate productivity sites (Appendix S3; Fig. 2). These differences correspond to the enhancement of species richness across the productivity gradient in these fields – some at low productivity and more at intermediate productivity sites. The fact that sorting was apparent after one growing season supports the view that establishment is a critical determinant of diversity patterns in grasslands (Houseman & Gross 2006). Interestingly, the number of species that showed a significant sorting pattern increased by the fourth growing season, suggesting that species traits – which influence growth and survival and are expressed later in the life cycle – will also be important in niche differentiation. These sorting patterns, coupled with an earlier experiment (Houseman & Gross 2006), support the view that differences in colonization along the gradient are driven by low moisture or nutrients at low productivity and competitive suppression at high productivity (Grime 1979).

We also tested whether species sorting could be linked to species traits that were independently measured in the greenhouse. Although the correlation coefficients from the FCA between PCA axes and productivity were not strong, there was a link between specific traits and the likelihood of initial establishment (Table 1). Early establishment along the gradient was associated with traits that correspond to seedling growth rate, seed size, germination characteristics and water use. Species with higher growth rates, larger and faster germinating seeds and high water use were more likely to colonize higher productivity sites. These trait-sorting relationships are consistent with physiological studies of plant responses to increased fertility (Lambers et al. 1998), suggesting a link between these physiological traits and successful colonization. Interestingly, in our study, species with higher root : shoot allocation were more likely to establish in high productivity sites. Often, root : shoot biomass decreases along productivity gradients (Zhou, Talley & Luo 2009), but increases have been reported (Snyman 2009). It may be that higher shoot allocation is required at the seedling stage to colonize productive sites because of low light levels beneath the canopy. By the fourth year, species associated with high growth rates and N-use were more successful as productivity increased but the other traits described above were no longer detectable. This is not surprising as our greenhouse assay of traits focused on early growth patterns. Similar reversals between seedling traits measured in the greenhouse and adult traits measured in the field have been reported in successional grasslands (Gleeson & Tilman 1994).

Other studies have examined the importance of species traits to species interactions (Grime 1977; Inouye & Tilman 1988; Keddy 1992; Latham 1992) and distributional patterns (Diaz, Cabido & Casanoves 1998; Eriksson & Jakobsson 1998; Reader 1998; Lavorel & Garnier 2002; Ozinga et al. 2004; Pakeman 2004; Lloret et al. 2005; Williams et al. 2005). However, many such studies often examine correlations between extant distributional patterns and environmental conditions. Although useful, such approaches cannot control for differences in species pool or immigration rates among environments. Because we manipulated seed inputs and independently assayed plant traits in the greenhouse, our results suggest that trait–environmental relationships contribute to the observed patterns of colonization across these communities and suggest that species sorting is important to patterns of diversity across resource gradients.

Niche vs. neutral colonization patterns

The debate over whether niche-based or neutral processes structure plant communities has relied largely on observational evidence comparing species composition across sites or habitats (Hubbell 2001; Bell 2005; Adler, HilleRisLambers & Levine 2007). There have been few experimental tests of processes that control colonization and determine diversity patterns (but see Fargione, Brown & Tilman 2003; Harpole & Tilman 2006). Our results suggest that, during colonization, species sort along broad environmental gradients and this sorting can be linked to specific traits as predicted by niche assembly theory. Clearly, there may be stochastic factors that influence local diversity at any site and contribute to community richness (Hubbell 2001; Tilman 2004). For example, stochastic mortality may reduce the advantage of competitive hierarchies during the establishment phase and allow weaker competitors to persist (Tilman 2004). This stochasticity is likely to magnify neutral assembly processes when regional immigration is highly variable.

Although we focused on productivity differences within sites, differences in colonization and establishment across sites may also depend on resident species composition (Fargione, Brown & Tilman 2003; Emery & Gross 2006). Our experiment cannot explicitly separate these two factors. We attempted to include fields with large differences in both productivity and community composition to maximize realistic variation of both factors (see Appendix S1) but cannot distinguish community composition from soil resources or other environmental factors as drivers of the ‘productivity’ response. However, despite substantial differences in species composition among fields, we found similar relationships between productivity and diversity following augmentation of species pools suggesting that as the size of the species pool increases, the underlying productivity–diversity pattern emerges (Houseman & Gross 2006; Zobel & Partel 2008).

Implications for metacommunities

Our results support the dual importance of regional (immigration) and local species interactions in determining species diversity that is inherent to metacommunity theory (Leibold et al. 2004). In these grasslands, species richness is limited by colonization (immigration) from the surrounding area and, when this limitation is released (by seed addition), colonization of successful immigrants at least partially reflects trait-based species sorting. The initial increase in plant species richness we observed in these fields after the seed augmentation persisted for 4 years illustrating that – even in systems with low disturbance – immigration may strongly influence community assembly and diversity.

To our knowledge, this is the first study to experimentally show that species pools can have consistent effects on local diversity across environmental gradients and that responses are at least partly due to trait-based species sorting occurring across environmental gradients during the early stages of plant establishment.

Acknowledgements

We thank Gary Mittelbach (GGM), Phil Robertson, Pete Murphy, Spencer Hall and Sarah Emery for providing valuable comments and feedback on earlier drafts of this article. Numerous high-school and undergraduate students from the Gross’ laboratory assisted with the field, greenhouse and laboratory research. The coordination and logistical support provided by Carol Baker was invaluable to the completion of the field and greenhouse studies. This research was supported by grants from the National Science Foundation, AW Mellon Foundation to KLG and GGM, and an NSF dissertation improvement grant to KLG and GRH and George H. Lauff Research awards to GRH. This is W.K. Kellogg Biological Station contribution number 1574.