Physiological drought tolerance and the structuring of tallgrass prairie assemblages

Drought is a defining characteristic of many grasslands worldwide. Yet we have little understanding of how drought structures grassland communities and the degree to which physiological drought tolerance advantages plants in grasslands. We characterized physiological drought tolerance (Wcrit) for a large number of species in a mesic grassland community (Konza Prairie, KS, USA). We then examined the relationships between Wcrit and a number of other key functional traits, and tested whether physiological tolerance of drought underlay success across a number of ecological contrasts—topographic position, burn frequency, and grazing—with 17 years of abundance data. Physiological drought tolerance of Konza species covered almost the full range known to plants globally. Consistently, physiologically drought-tolerant species had thin roots, while associations with other traits were inconsistent across functional groups. In this mesic grassland, physiological drought tolerance appears to increase the abundance of plants in xeric uplands, but does not in the mesic lowlands. Physiological drought tolerance did not alter species responses to changes in burning or grazing. In contrast to Wcrit, species with high root tissue density were more abundant in uplands and lowlands than species with low root tissue density largely irrespective of grazing or burning regimes. In all, drought appears to have a limited role in structuring the Konza plant community. As such, more severe or frequent droughts in the region would likely restructure the Konza plant community in ways that are currently not observable.


INTRODUCTION
The episodic nature of precipitation produces moisture stress at multiple scales, from decadelong reductions in precipitation, seasonal dry periods, and daily mid-day inductions of plant water stress (Weaver 1968, Karl and Koscielny 1982, Knapp 1985, Woodhouse and Overpeck 1998. Tropical and Mediterranean grasslands typically have annual dry seasons during which grasses senesce (Gibson 2009). Temperate grasslands are not characterized by a dry dormant season, but periodically experience years with low precipitation as well as soil moisture stress on shorter time scales (Gibson 2009). Across a wide variety of temperate grassland ecosystems, periods of low soil moisture, i.e., drought, reduce productivity (Frank and McNaughton 1992, Briggs and Knapp 1995, Ciais et al. 2005) and alter species abundance (Weaver 1968, Tilman andEl Haddi 1992). With projected increases in temperature and reduced water availability during the growing season for temperate grasslands (Christensen et al. 2007), drought is likely to increase in importance as a driver of future grassland community structure.
To predict how grasslands will be structured by drought requires not only forecasts of future climates, but understanding of how drought currently structures communities. One inferential approach is to examine the relationships between species abundance and traits that confer tolerance of drought. Although there are a number of strategies associated with drought (McDowell et al. 2008, Craine 2009), places where soil moisture stress is most frequent and intense tend to be dominated by species that can photosynthesize at low plant water potentials without xylem cavitation (Pockman and Sperry 2000, Bhaskar and Ackerly 2006, Grime et al. 2008, McDowell et al. 2008. This physiological drought tolerance requires sclerification of vascular bundles and small vessel diameter to resist the physical stresses associated with low water potential (Hacke et al. 2001, Sperry et al. 2003. Indirectly, physiological drought tolerance is expected to be linked to other functional traits (Ackerly 2004). For example, stress-tolerant species should have low rates of gas exchange and low maximal growth rates due to inherent tradeoffs in xylem functioning (Grime et al. 1997, Reich et al. 2003a). Furthermore, physiologically drought tolerant species should, all other things equal, have higher leaf and root tissue density resulting from the requirement for increased cell wall investment (Cunningham et al. 1999). Thin leaves (Niinemets 2001) and high leaf angle (Ehleringer 1983, Medina et al. 1990, Craine et al. 2002 are also likely associated with drought tolerance. Despite the potential of drought to structure temperate grassland communities and historic changes in communities, tests of the role of drought-associated traits in determining abundance in grasslands are rare. For example, at Konza Prairie, a North American tallgrass prairie, interannual variation in precipitation drives variation in productivity and flowering (Briggs and Knapp 1995, Nippert et al. 2006), but seasonal and spatial gradients result in low soil moisture availability even in years with above-average precipitation (Nippert et al. 2011). Topographic position, nutrient availability, fire, and grazing all strongly influence community structure at Konza (Collins et al. 1998, Towne et al. 2005, Spasojevic et al. 2010), but it is unknown whether drought also has a strong influence on community composi-tion at Konza or whether these factors alter the importance of soil moisture stress. At Konza, leaf tissue density from field-collected plants consistently correlated with abundance (Craine and Towne 2010) across a number of ecological contrasts. High leaf tissue density could indicate greater drought tolerance, but also is known to be associated with success in low-nutrient environments (Craine et al. 2001) while providing resistance to herbivory (Wright andIllius 1995, Perez-Harguindeguy et al. 2003). In an irrigation experiment at Konza, dry soils favored species with low N concentrations and high leaf longevity, which is the opposite of geographic patterns where low-precipitation sites tend to be dominated by plants with higher leaf N concentrations and low leaf longevity (Sandel et al. 2010).
To better understand patterns of physiological drought tolerance among grassland species and the role it plays in determining abundance in a mesic grassland, we first measured physiological drought tolerance and a corresponding set of functional traits for a broad suite of grassland species present at a mesic prairie (Konza Prairie) in central North America. We then tested the ability of physiological drought tolerance, other individual traits, and correlated suites of traits to predict the abundance of species across a number of ecological contrasts in a tallgrass prairie. These included xeric uplands and mesic lowlands, frequently and infrequently burned areas, as well as grazed and ungrazed areas.
Regarding the suites of traits that might be associated with physiological drought tolerance, if physiological drought tolerance is dependent on additional allocation to cell walls, then physiologically drought-tolerant species should have high leaf and root tissue density and lower rates of gas exchange. Regarding the role of drought in determining abundance, if periodic or chronically low soil moisture structures the grassland plant communities and plants are reduced in abundance by the physical stresses associated with low soil moisture, then species with greater physiological drought tolerance should be more abundant. As topographic variation can alter plant responses to climate, the importance of drought tolerance is unlikely to be uniform across the landscape , Debinski et al. 2010. Shallow-soil uplands, burned areas, and ungrazed sites at Konza are the most likely to experience seasonal soil moisture stress (Knapp et al. 1993, Fahnestock and Knapp 1994, Blair 1997, Bremer et al. 2001, Nippert and Knapp 2007b and therefore are the most likely to be dominated by physiologically drought tolerant species. Alternatively, if drought is not an important structuring factor for the grassland communities, or if drought is severe enough to cause mortality for even droughttolerant species, there might be no relationship or even a negative relationship between physiological drought tolerance and abundance.

Site description
Konza Prairie Biological Station is a 3487 ha native tallgrass prairie located in the Flint Hills of northeastern Kansas, USA (39805 0 N, 96835 0 W). The prairie landscape is dominated by a few species of warm-season grasses (Andropogon gerardii, Sorghastrum nutans, Schizachyrium scoparium, and Panicum virgatum) while cool-season grasses and a diverse suite of forbs comprise a large proportion of the rest of the plant community. Konza receives an average of 835 mm of precipitation annually, most of which (75%) falls during the growing season. Over the last century at Konza, mean annual precipitation regularly deviated from the long term mean by about 25% and reached values as high as 184% of the mean in the wettest year (1533 mm in 1951) and 47% in the driest year (392 mm in 1966). While the mean annual temperature for Konza is 138C, the mean low for the year is À38C in January and the mean high of 278C occurs in July.

Plant cultivation
One hundred and twenty-one herbaceous tallgrass prairie species were chosen for this study. Species chosen for the experiment encompass a broad range of attributes. Phylogeny, life history, and seed availability were all considered during the selection process in order to best represent the floral diversity found on Konza. Seeds were obtained from a variety of sources, including the Kansas Native Plant Society, the National Plant Germplasm System, Chicago Botanic Garden-National Tallgrass Prairie Seed Bank, Taylor Seed Farms (White Cloud, KS), and local collection by the authors at Konza.
Propagules were germinated on damp filter paper in Petri plates at room temperature. Stratified seeds were stored on damp filter paper in a 58C incubator for at least 30 days while those that required scarification were abraded with sandpaper before being germinated in appropriate conditions. Seedlings were transplanted to 164 mL plastic Cone-tainers (D-40, Stuewe and Sons, Inc. Corvalis, OR) containing standardized, untreated lowland soil from Konza (silty clay loam). Similar to previous experiments (Reich et al. 2003b), plants in containers were grown in a Conviron growth chamber (Model PGV 36, Controlled Environments Limited, Winnipeg, Manitoba) with 16-hour days at 258C and light levels at 1200 lmol m À2 s À1 . Temperatures at night were held at 208C. Plants were watered daily and treated with a commercial fertilizer (Miracle Grow 24-8-16 All Purpose Fertilizer) biweekly to eliminate nutrient stress. An average of 8 replicates of each species were maintained, with the level of replication varying from 1 to 24 plants.

Trait measurements
Plants were grown in the growth chamber for 8-12 weeks before data collection. Gas exchange was measured using a Li-6400 infra-red gas analyzer with red/blue LED light source and CO 2 injector (LICOR Biosciences, Lincoln, NE). Light intensity inside the cuvette was 2000 lmol m À2 s À1 , CO 2 concentration was 400 ppm, and relative humidity was 40%. Measurements were performed on the newest fully-expanded leaves and included maximum photosynthetic rate (A max ), stomatal conductance to vapor (g s ), and water use efficiency (WUE; A max /transpiration rate).
Leaf thickness (Thick L ) was measured between secondary veins where applicable for 2-3 newlyexpanded, mature leaves on each plant using digital calipers (Thermo Fisher Scientific Inc., Waltham, MA). Leaf angle (Angle L ) relative to horizontal was measured by averaging 3-5 protractor measurements per plant (Craine et al. 2001).
After 8-12 weeks, plants were divided into sets of 50 and were subjected to a dry-down period with daily monitoring using a steady-state diffusion porometer (Model SC-1, Decagon Devices, Inc., Pullman, WA). Stomatal conductance was recorded daily during dry-down until the conductance rate fell below 5% of the maximum. Following stomatal closure, non-senesced leaf tissue was collected and the hydrostatic pressure potential was measured using a Scholander pressure bomb (PMS Instrument Company, Albany, OR). This leaf pressure potential corresponding to stomatal closure is henceforth referred to as the species' critical water potential (W crit ), and is used in the study as an index of physiological drought tolerance. A subset of the leaves was used to measure leaf area (LI-COR Leaf Area Meter, Model LI-3100) and subsequently oven dried and weighed to calculate specific leaf area (SLA).
Leaf tissue density (q L ), the ratio of leaf mass to leaf volume was calculated using leaf area and thickness. The remaining biomass was sorted to leaf or stem and dried at 608C to determine total aboveground biomass (M L ). Roots were sorted into coarse (.2 mm) and fine roots. A representative sample of the fine roots was scanned into a digital root imaging program (Winrhizo; Regent Instruments, Inc., Nepean, Ontario, Canada) which calculated total root length, total root volume, and average root diameter by length. The remainder of the roots was oven dried, weighed, and used to calculate specific root length (SRL), root tissue density (q R ), average root diameter (Diam R ) and total root biomass (M R ).

Species abundance
As part of the long-term research at Konza, plant composition has been sampled twice annually (late May-June and mid-August-September) since 1983, to capture canopy cover for both early-and late-season species. Twenty watersheds were chosen  to represent the six land-management treatments: two topographic positions within grazed and ungrazed watersheds with burn intervals of 1 or 20 years. In each watershed there were eight 50-m permanent transects. Half of the transects were located in shallow, rocky upland soil while the other half were located in deep, fertile lowland soil. Relative cover was sampled in five permanently-marked circular plots (10 m 2 ) that were evenly spaced along each transect. A modified Daubenmeier cover scale (Bailey and Poulton 1968) was used to visually estimate species cover , which is used as the index of abundance here.
Average relative abundance in the watershed for each year was calculated by selecting the larger cover measurement for each species from the two sample periods and using the midpoint of the cover class to average across all upland or lowland plots in the watershed. We averaged across 17 years (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) to yield a single relative abundance value for each prairie species. Relative abundance for each watershed was combined across similar treatments to gain average values for each treatment. For example, abundance was averaged across all grazed watersheds to gain a single value of relative abundance for the grazing treatment that could be compared to the ungrazed treatment. This was done for grazed, ungrazed, annual burns (burned), 20 year burns (unburned), upland, and lowland treatments. These categories are referred to as contrasts. Finally, we averaged across all treatments to get a single average relative abundance value for each species.

Statistical analysis
Ten functional traits were chosen as the primary functional traits of interest. Encompassing tissue and whole plant morphology and physiology, these traits included root and leaf tissue density, average root diameter, leaf thickness, leaf angle, root mass, shoot mass, W crit , maximum photosynthetic rates and maximum conductance rates. The 10 traits were used in pairwise correlations and in principal component analyses performed on correlations (JMP 8.0.2, SAS Institute, Inc., Cary, NC). Principal component analysis was performed with these 10 traits for the entire species set and then for grasses alone and forbs alone. For the abundance measures, we only examined single-factor contrasts. After determining the average abundance of species for an ecological contrast, e.g., uplands or unburned watersheds, all abundance data were log-transformed to normalize abundance data distributions. We tested whether traits predicted the change in abundance of species across the ecological contrast, which was defined as the difference in log-transformed abundance for a species between the two contrasts. With the abundance data log-transformed, this difference represents the relative change in species abunv www.esajournals.org dance independent of its abundance in either contrast. To test which traits best predicted abundance, stepwise multiple linear regressions were performed for each contrast and the most parsimonious model was selected using AIC scores. We tested whether relationships between W crit and abundance in uplands and lowlands differed among years with a repeated measures ANOVA.

Trait relationships
Among species, W crit ranged from À1.1 MPa in Tradescantia bracteata to À8.9 MPa in Bouteloua curtipendula. Seven species had W crit ! À2.0 MPa while 11 species had W crit À8.0 MPa. The grasses that we measured were generally more physiologically tolerant of drought than forbs (À6.8 vs. À4.0 MPa, respectively; P , 0.001). Yet, there were several forbs that were as tolerant as the most tolerant grasses, e.g., Salvia reflexa (W crit ¼ À8.6 MPa), and some grasses such as Panicum virgatum and Eleusine indica had low physiological drought tolerance (.À3.5 MPa). Annuals were not more drought tolerant than perennials (P ¼ 0.18), nor were C 4 species more drought tolerant than C 3 species (P ¼ 0.32). Among the other morphological traits, Xanthium strumarium had the least dense leaves (0.10 g cm À3 ) while Andropogon gerardii had the most dense (0.86 g cm À3 ). The thinnest leaves belonged to Chloris verticillata (0.08 mm) while Silphium lacinatum had the thickest (0.57 mm). Root tissue density (q R ) ranged from 0.11 g cm À3 in Euphorbia marginata to 0.58 g cm À3 in Amorpha cansecens while average root diameter varied from 0.09 mm (Lepidium densiflorum) to 0.45 mm (Xanthium strumarium).
Among the 10 main functional traits, 47% of the pairwise correlations were significant ( Table  1). The strongest correlation was between the two gas exchange variables as species with high photosynthetic rates had the highest stomatal conductance (r ¼ 0.70, P , 0.001). Overall, W crit correlated with 4 of the 9 other main functional traits. Species that were more physiologically tolerant of drought (lower W crit ) had thin leaves (r ¼ 0.28, P , 0.001) that were dense (r ¼À0.37, P , 0.001) and held at a high angle (r ¼À0.39, P , 0.001) as well as thin roots (r ¼ 0.54, P , 0.001) ( Fig. 1). Root tissue density was not directly correlated with W crit (P ¼ 0.09), nor were the two gas exchange metrics or biomass amounts.
In the PCA of the 10 main functional traits, the first trait axis separated drought-tolerant species from drought-intolerant species (Table 2). Six traits contributed significantly to the axis, accounting for 28.3% of the total variation in all traits among all species explained by PCA. Species that were physiologically tolerant of drought had thin, dense leaves that were held at a high angle and thin, dense roots. For example, Hesperostipa spartea which continued stomatal conductance to À8.0 MPa had leaves that were 0.62 g cm À3 and only 0.12 mm thick. In contrast, Asclepias speciosa ceased stomatal conductance at À2.0 MPa. Its leaves had a density of only 0.27 g cm À3 and were 0.19 mm thick.
On average, grasses had a more droughttolerant strategy than forbs (P , 0.001) and a simple dichotomy of species into grasses and forbs explained 50% of the variation in Axis 1 (Table 3). Neither photosynthetic pathway nor life history was associated with differences in Axis 1 (Table 3). With differences in grasses and forbs explaining a large proportion of the variation in Axis 1, multivariate analyses for the 10 main functional traits were run separately for the two groups. Although W crit was still associated with other functional traits for both grasses and forbs, root diameter and low rates of photosynthesis were the only traits that were consistently associated with W crit (Table 2). For forbs, physiologically drought tolerant species had thin, dense roots and low rates of gas exchange. For grasses, physiologically drought tolerant species had thin dense leaves held at a high angle, thin roots, and a low maximal rate of photosynthesis. Photosynthetic rate was not, however, correlated with physiological drought tolerance (r ¼À0.01, P ¼ 0.9) nor was it associated with W crit in the broader principal component analysis.
Axis 2 reflected the correlation among species in gas exchange rates that were largely independent of drought-tolerance ( Table 2). As seen in the bivariate relationships, species with high photosynthetic rates also had high rates of stomatal conductance and their leaves were held at a high angle. These species also had a higher fraction of root biomass than those low on the v www.esajournals.org axis (r ¼ 0.32, P , 0.001). On average forbs scored lower than grasses on Axis 2, which reflects their lower rates of gas exchange (Table 3). The third axis primarily separated species based on their size at the end of the experiment and only explained 3.7% more variation than expected by chance (Table 2).

Species abundance
Species that were physiologically tolerant of drought were not more abundant overall at Konza (P ¼ 0.12; Fig. 2). Physiologically drought-tolerant species were more abundant in uplands (y ¼ À2.20 À 0.22 3 W crit ; r 2 ¼ 0.12, P ¼ 0.008) and infrequently burned areas (y ¼À2.55 À 0.19 3 W crit ; r 2 ¼ 0.09, P ¼ 0.01), while physiologically drought-tolerant species had greater differential abundance in uplands vs. lowlands (Abund Upl À Abund Lowl ¼À1.00 À 0.20 3 W crit ; r 2 ¼ 0.18, P ¼ 0.001). The relationships between W crit and species abundance in the uplands did not vary over time (P . 0.68). Physiological drought tolerance did not explain a significant amount of variation in average species abundance for lowlands, frequently burned, grazed, or ungrazed areas, with this lack of relationship consistent across years (P . 0.5). Physiological drought tolerance also did not predict the response of species to burning or grazing (P . 0.9). There were no relationships between variation in abundance over time for species (here, coefficient of variation for untransformed abundance) and W crit for either uplands (P . 0.37) or lowlands (P . 0.85).
Of the 9 other metrics measured for each species, root tissue density was consistently the best predictor of abundance (Fig. 3). High root tissue density species were more abundant overall (r 2 ¼ 0.12, P ¼ 0.003) and were more abundant in each of the six ecological contrasts (Fig. 3). The relative abundance of high root tissue density species did not change any more than low tissue density species when burned or grazed, nor were high root tissue density species relatively more abundant in uplands than lowlands (P . 0.05 for all 3 contrasts). Of the remaining 80 combinations of 8 traits and 10 abundance metrics, only 11 were significant. The strongest relationship was between leaf angle and abundance in grazed areas (r 2 ¼ 0.14, P ¼ 0.002). Axis 1 scores explained just 6.7% of the variation in overall abundance (P ¼ 0.02) and it v www.esajournals.org consistently explained less variation than individual traits for the other contrasts (data not shown). Axes 2 and 3 explained no significant variation in abundance overall or for any contrast.
Using backwards-elimination linear regressions to determine the best predictors of abundance, W crit was the best predictor of abundance for uplands and for the difference in abundance between uplands and lowlands (Table 4). In many other cases, root tissue density was the sole predictor of abundance or contributed to the best predictive model (Table 4). For overall abundance, root tissue density and leaf angle jointly described 19% of the variation in abundance. The amount of variation in abundance that was explained by combinations of traits ranged from 8% to 23%. 29% of the difference in abundance between uplands and lowlands was explained by a combination of W crit and leaf tissue density (Table 4).

DISCUSSION
In the xeric uplands at Konza, soil moisture stress appears to be frequent enough and/or sufficiently intense to have favored plant species with greater physiological drought tolerance. Although W crit only explained ;12% of the variation in abundance, the most physiologically drought-tolerant species was predicted to be 50 times more abundant than the least physiologically drought tolerant species. Compared to the over 100,000-fold variation in upland abundance among all species, the amount of variation explained by drought tolerance might appear small, but on an absolute scale it is sizeable nonetheless. Konza upland soils tend to be Notes: Axes were rotated to strengthen contrasts among axes. Bold values represent variables that contribute significantly to a given axis (absolute value . 0.33 is equivalent to P , 0.05). Eigenvalues represent the ratio of variation explained by a given axis to the amount of variation explained by chance. Note: For each contrast we report least squares means (LSM) and partial r 2 , each contrast's proportion of the total variation explained by the model (a ¼ 0.05).
v www.esajournals.org lowland community in the past or in a subset of years. Droughts also might have restricted species presence in the lowlands, filtering sensitive species out of the current community. Only 240 of Konza's ;550 herbaceous species have been observed in the lowland composition plots over the 17 years of monitoring; so historic drought may be responsible for some of the reduction in the observed species set. That said, there was no average difference in W crit between species found in lowland plots and those not found in upland or lowland plots (P ¼ 0.85). Alternatively, species with high W crit might possess other adaptations that allow them to recover quickly after drought. If so, drought might structure the community beyond selecting for species with low W crit . At present, there is no evidence that drought has structured the lowland communities in ways that favor physiologically drought-tolerant species as observed for uplands. Plant water potentials are generally higher in the lowlands than uplands at Konza, which supports the conclusion of low importance of drought as a structuring agent in lowland communities. The general pattern of greater drought stress in the uplands than the lowlands is congruent with patterns observed for a grassland in northern England (Buckland et al. 1997). There, shallow soils had much lower soil water potentials during an extreme drought and plants with low physiological drought tolerance showed greater reductions in leaf water than drought-tolerant species.
Regardless of the benefit of drought tolerance in the uplands, for both uplands and lowlands, many physiologically drought-intolerant species were still abundant. Some drought-intolerant species escape drought spatially by occupying microsites where drought is less important (Buckland et al. 1997). Others escape drought phenologically, restricting most of their growth to the wetter, milder spring and fall seasons (Taiz and Zeiger 2002). Annuals can complete their life cycle before severe drought occurs, while some perennial species senesce before drought becomes too severe (Heckathorn and Delucia 1996). Lastly, many of the species with high W crit , such as Lespedeza capitata, avoid drought stress by accessing deep soil water (Canadell et al. 1996, Buckland et al. 1997. Soil water is relatively available at depths greater than 1 meter Knapp 1995, Nippert andKnapp 2007b) on Konza regardless of antecedent precipitation patterns. Future research that quantifies more specific habitat preferences and/or other related functional traits should shed light into the Notes: Abundance data tested also included differences in abundance between uplands and lowlands (DUpland), burned and unburned watersheds (DBurn), and grazed and ungrazed watersheds (DGraze). Partial r 2 is the proportion of model r 2 contributed by each trait. Estimate is the slope of the relationship between the trait and the abundance metric. Sum of Squares represented by SS.
v www.esajournals.org maintenance of diversity at Konza.
Irrespective of its role in structuring the community, physiological drought tolerance does not appear to be part of an obvious broader plant strategy among Konza species. Outside of thin roots, physiological drought tolerance was not consistently associated with specific functional traits as correlations depended on the groups examined. In addition, the correlated suite of traits that emerged from grasses and forbs predicted less variation in abundance than W crit in the uplands. Still, low W crit and thin roots were consistently associated with one another. The consistent association between low W crit and thin roots was not just a consequence of grasses having thinner roots than forbs for the species studied here as the association held within grasses and within forbs. The low average root diameter of physiologically drought-tolerant species might be a product of thinner xylem elements to prevent cavitation even under very high tension, but could also be associated with a reduced cortex. Distally, the thin roots could represent selection directly related to resistance to low soil water potential, but also could represent selection related to changes in the relative importance of water and nitrogen. A more detailed analysis of the anatomical determinants of interspecific differences in diameter along with models of water and nutrient movement in soils are needed to begin to understand the association between physiological drought tolerance and thin roots.
With our results, one of the key questions to begin to answer is not whether or not drought structures Konza plant communities, but what is the relative role of drought compared to other stresses and disturbances. To this point, root tissue density was consistently associated with abundance in uplands and lowlands, explaining more than twice the variation in abundance than W crit on average. There was no correlation between W crit and root tissue density which implies that high root tissue density is not necessarily a determinant of physiological drought tolerance. High root tissue density has often been linked with low-nutrient environments and tends to be correlated with high leaf tissue density (Wahl and Ryser 2000). Dense roots have low turnover rates (Craine et al. 2002) and are robust due to a high percentage of root stele (Wahl and Ryser 2000). High root tissue density also imparts success when nutrients are limiting. For example, the high root longevity, to which high root tissue density contributes, generates high root length density in soil, which confers the ability to preempt nutrient supplies from competitors (Craine et al. 2005). If high root tissue density is the best trait to indicate importance of low nutrient availability, it is possible that nutrient limitation is more important than drought in structuring the Konza plant community in the lowlands, and of similar importance in uplands. At Konza, both upland and lowland productivity can be limited by N , which would support the importance of nutrients in both landscape positions.
Ultimately, understanding the relative role of drought in structuring the plant communities of grasslands will require monitoring of plant communities during and after severe droughts to determine the immediate stress effects and the community's recovery. At Konza, severe droughts have not occurred since the early 1980s, and for many years the leaf water potential of the dominant grasses has not declined below À3 MPa. As such, although landscape-level variation in soil moisture might structure the grassland community (Silvertown et al. 1999), it is likely that in the more mesic areas at Konza, drought has been less important than other environmental factors such as low nutrient availability, fire, and grazing. If drought becomes more frequent or severe in the future, plant communities might increasingly resemble the uplands and species with low W crit might become more abundant.

ACKNOWLEDGMENTS
The Konza Prairie LTER data set analyzed here is PVC02, which was collected by Gene Towne during the 17 years for which we analyzed. Teall Culbertson, Gracie Orozco, and Troy Ocheltree assisted in collection of trait data. We appreciate the provision of seeds from the many sources described earlier. Data collection and archival work was supported by a National Science Foundation grant to the Konza Prairie LTER program (DEB-0823341