Linking abundances of the dung fungus Sporormiella to the density of bison: implications for assessing grazing by megaherbivores in palaeorecords

Authors


Correspondence author. E-mail: jacquelynlgill@gmail.com

Summary

  1. Megaherbivores likely had important influences on past vegetation dynamics, just as they do in modern ecosystems. The exact nature of megaherbivores' role can be studied using a relatively new suite of palaeoecological techniques, including the quantification of fossil spores from Sporormiella and other coprophilous fungi as indicators of megafaunal biomass in sediment records. However, a quantitative linkage of spore abundance with megaherbivore biomass or grazing intensity has been lacking.

  2. Konza Prairie Biological Station (Kansas, USA), located in the midcontinent of North America, contains native tallgrass prairie grazed by a herd of bison (Bison bison) in a 1000-ha enclosure, providing an excellent opportunity to test the effects of megaherbivores on grassland community composition and their potential signature in the palynological record. We collected pollen and spores during 2009 and 2010 from a network of 28 modified Tauber traps. The precise locations of the bison herd were recorded using GPS collars; we calculated bison grazing intensity (kg m−2 year−1) to high spatial precision within concentric circles around each trap (radii from 25 to 500 m).

  3. Both relative (per cent) and absolute (concentration) abundances of Sporormiella were significantly higher in traps inside the enclosure and were positively correlated with bison grazing intensity. The cut-off for distinguishing between bison-grazed and ungrazed traps was determined to be 2.8% Sporormiella of the total pollen and spore sum, consistent with previous palaeoecological reconstructions. The relationship between Sporormiella abundances and available grazing area around each trap was strongest at short radii (25–100 m), suggesting that spores do not disperse far from their source. Sporormiella should thus be considered a local-scale indicator of megaherbivore presence.

  4. Traps in the grazed area had significantly higher percentages of Ambrosia and lower percentages of Poaceae pollen than traps from ungrazed areas. This suggests that the pollen record has the potential to detect the ecological effects of bison grazing on grassland community composition.

  5. Synthesis. This study refines the use of Sporormiella as a proxy for local megaherbivore presence, especially in grassland systems. Multiproxy Sporormiella and pollen analyses may help elucidate the past drivers of grassland dynamics, including the possible role of bison in mediating grass–forb interactions during the variable moisture regimes of the last 12,000 years.

Introduction

The global wave of megaherbivore extinctions during the late Pleistocene and Holocene represents one of the major ecological upheavals of the recent past. There has been a growing interest in understanding the cascading effects of these extinctions despite the challenge that the signature of herbivory and other species interactions is usually indirect in palaeorecords (Johnson 2009). Our understanding of megaherbivory as an ecological process is mostly based on contemporary experimental manipulations (Hester et al. 2000; Asner et al. 2009) or the post hoc assessment of the effects of local extirpations or management practices (Constible et al. 2005; ter Beest 2006). Recently, the use of coprophilous fungi as indicators of palaeoherbivory in sedimentary archives has led to significant advances in our understanding of the ecological legacy – and consequences of the loss – of megaherbivores throughout the late Quaternary. In particular, declining abundances of spores from the genus Sporormiella (Davis 1987; van Geel 2002; Davis & Shafer 2006) in sedimentary records has been used to link megafaunal population declines and extinction to changes in plant community composition and fire regimes in New York (Robinson, Burney & Burney 2005) and late Holocene changes in Madagascar (Burney, Robinson & Burney 2003), as well as the formation of novel plant associations in the Great Lakes region of North America (Gill et al. 2009, 2012) and shifts in biomes in Australia (Rule et al. 2012) during the late Pleistocene. However, in spite of its increasing use in a number of high-profile palaeoecological studies, little is known about the natural history of Sporormiella, including palaeohost preferences, production, dispersal and taphonomy (Feranec et al. 2011). Of particular importance is quantifying the relationship between spore abundance and megafaunal biomass or herbivory intensity. Jackson (2012) called for the palaeoecological community to avoid ‘ignorance creep’ by addressing the conceptual gaps linking target variables (e.g. herbivore grazing) with proxy data (e.g. Sporormiella). In this paper, we resolve several key ‘known unknowns’ (Jackson 2012) about the use of the Sporormiella proxy in palaeorecords, making use of the bison enclosure experiment at the Konza Prairie Biological Station to provide a uniquely detailed analysis of the relationship between Sporormiella and megaherbivory.

Sporormiella has potentially high utility as a palaeoenvironmental proxy because, unlike most other species of coprophilous fungi, it is both obligate to herbivory and produces distinctive spores that preserve well in the fossil record. Spores must pass through the digestive system of a herbivore in order to reproduce, and the fungus sporulates on a dung substrate. New spores are passively dispersed to the surrounding vegetation, where they may be consumed by herbivores to continue the fungus' life cycle (Richardson 2001; Nyberg & Persson 2002). As a non-pollen palynomorph, Sporormiella is complementary to the vertebrate bone record, which offers direct but discontinuous evidence of megafaunal presence. Unlike vertebrate remains, Sporormiella spores (i) are best preserved in lake and mire sediments, which are often temporally continuous, (ii) can be counted for estimates of relative and absolute abundance (like pollen and other microfossils) and (iii) are found in the same depositional environments as fossil pollen and other palaeoecological and palaeoclimatic proxies. Sporormiella has many modern hosts (Ahmed & Cain 1972), but its signal in late Quaternary records appears to be strongly linked to megafaunal population abundances, presumably because megaherbivores produced such great quantities of dung that the spores are sufficiently abundant in sediment records. In North America, Sporormiella spores are abundant in mammoth coprolites and late Pleistocene sediments (> 11 700 BP; Davis 1987; Robinson, Burney & Burney 2005; Gill et al. 2009, 2012) and are rare (usually < 2% of the upland pollen sum) in Holocene (< 11 700 BP) lake and bog records. Sporormiella abundances in sediments often increase in association with the historic introduction of cattle, horses and other domesticated herbivores (Robinson, Burney & Burney 2005; Davis & Shafer 2006). Previous studies have shown a positive relationship between the abundance of Sporormiella in surface sediments and the historic presence of cattle and other large grazers (Graf & Chmura 2006; Raper & Bush 2009; Parker & Williams 2012; Etienne et al. 2013). However, these analyses did not quantify the source area of spores and were limited by a lack of detailed data on megaherbivore densities. Therefore, the quantitative relationship between Sporormiella abundances, megaherbivore biomass and grazing intensity has remained less well known, although it is of critical importance in the interpretation of the Sporormiella proxy.

Bison (Bison bison; hereafter ‘bison’) are one of only 10 species of megaherbivores to survive the late Quaternary extinctions in North America, of the 45 genera present on the continent at the end of the Pleistocene (Koch & Barnosky 2006; Faith & Surovell 2009). Bison are considered a keystone species of the tallgrass prairie for their well-documented influences on grassland structure, composition and ecosystem function (Knapp et al. 1999). The late Holocene near extirpation of bison is thought to have significantly influenced the ecology of the modern Great Plains, and bison grazing, along with fire, has become an important component of prairie restoration efforts (Hartnett, Hickman & Fischer Walter 1996; Collins et al. 1998; Martin & Wilsey 2006; Fuhlendorf et al. 2008). The Konza Prairie Biological Station (39.08° N, 96.58° W, mean elevation 397 m) located in the Flint Hills of Kansas, USA (Fig. 1), provides an opportunity to test the relationship between Sporormiella abundance and megaherbivory. Konza contains a well-studied herd of free-roaming bison in a 1000-ha enclosure in the tallgrass prairie biome. Annual records of bison biomass have been maintained since 1994, and bison position within the landscape has been monitored via GPS collars worn by matriarchal females. Therefore, in addition to testing the megaherbivore–Sporormiella relationship, in this study, we also assess the ability of the pollen record to detect the effects of bison grazing on plant community composition.

Figure 1.

Map of Konza and experimental design. The bison enclosure is indicated by the green line and shading. Cattle-grazed area (tan shading) doubled between 2009 (brown hatches) and 2010. Numbered trap locations are indicated with orange dots. For traps near watershed divides, the position of the number indicates which watershed the trap is located within. The ring distances used with the bison telemetry data are shown for trap 759 as an example. The same colour scheme is used to indicate ring distances in Figs 4 and 5.

Materials and methods

Site Description and Context

The Konza Prairie Biological Station (hereafter ‘Konza’) is a 3487-ha native tallgrass prairie remnant (Fig. 1) on the eastern edge of the Great Plains of North America. Located in the Flint Hills ecoregion, Konza has been protected from agricultural use by the region's shallow, rocky limestone soils and relatively high topographic relief. Since 1981, a series of watershed-level experiments with several replicated grazing and prescribed fire treatments have been funded by the National Science Foundation's Long-Term Ecological Research Program. In 1987, free-roaming native bison were introduced to the site. Today, the herd consists of approximately 300 individuals present throughout the year in a 1000-ha enclosure (Fig. 1). Supplemental feed is not provided to this herd; the bison stocking rate is designed to remove 25% of above-ground annual primary productivity. Grazing cattle are seasonally present in some watersheds at Konza (separate from the bison enclosure); grazing treatments changed in 2010 to include more cattle-grazed watersheds (Fig. 1). Bison are the dominant megaherbivore within the enclosure, although white-tailed deer (Odocoileus virginianus) are occasionally present. The vegetation at Konza is predominately C4 perennial grasses, including big bluestem (Andropogon gerardii), indiangrass (Sorghastrum nutans), little bluestem (Schizachyrium scoparius) and switchgrass (Pancium virgatum), along with almost 600 species of subdominant C3 and C4 graminoids, forbs and woody plants (Freeman & Hulbert 1985). Gallery forests consisting of oak (Quercus spp.) and the common hackberry (Celtis occidentalis) are common near streams (Abrams 1986). Vegetation at Konza is highly structured by bison and fire treatments (Fig. S1b-c in Supporting Information); both grazing and periodic fire are necessary to maximize diversity and community stability (Collins & Calabrese 2012).

Pollen and Spore Data

Twenty-eight modified Tauber pollen traps (Fig. S1a) were deployed at the end of the October 2008 flowering season (Tauber 1974; Hicks & Hyvarinen 1986) both inside and outside the bison enclosure (Fig. 1). Tauber traps are thought to collect pollen and spores from a very local source area, transported primarily from the air, as opposed to other transport mechanisms such as slopewash (Tauber 1974). Trap deployment locations were determined by a randomizing GIS algorithm balancing accessibility (e.g. proximity to roads), level soil surface (so as not to bias the dispersal of pollen and spores into the trap), equal coverage of grazed and ungrazed locations and even distribution throughout the watersheds. Traps were placed with at least 500 m ground distance between them. Trap contents were collected in October 2009 and 2010. Many of the 2009 traps were found to contain large amounts of fungal growth, dung beetles that had fallen into the traps or both. The 2010 traps were therefore modified to include a coarse mesh covering (to prevent beetle entry but not to restrict air flow and pollen transport) and 2 g of the fungicide thymol to prevent the growth of fungus within the trap (this fungicide did not affect the preservation of fungal spores transported into the trap). Due to the high volume of contaminants in the 2009 traps, only 11 traps had contents suitable for pollen and spore analyses. All 28 of the 2010 traps were collected and analysed. Cattle grazing treatments changed in 2010 to include more watersheds, and so, three of the 2010 traps (710, 763 and 764) were located inside seasonally grazed cattle watersheds for which stocking densities and grazing intensities were unavailable (Fig. 1). These were designated as ‘cattle’ samples and were excluded from most analyses, except where explicitly stated otherwise. In addition to the traps, eight samples from fresh bison dung were processed using standard pollen preparation methods (Faegri & Iverson 1989) to determine that bison dung contained Sporormiella spores, which to our knowledge has not been previously reported in the coprophilous fungus literature.

Trap contents were collected in the field by rinsing each trap with deionized (DI) water into a clean plastic gallon jug. Jug contents were passed through a cellulose filter; the filter and captured material were soaked with glacial acetic acid to prevent the growth of mould or fungi. Samples were processed using standard methods for pollen analysis, modified to include a method for dissolving the filter based on the European Pollen Monitoring Programme (Faegri & Iverson 1989). Each filter was soaked in 15 mL each of glacial acetic acid and acetic anhydride for 5 min. A mixture of 45 mL of acetic anhydride, 15 mL of glacial acetic acid and 7 mL of concentrated sulphuric acid was added to each sample to dissolve the cellulose filters. Samples were condensed and rinsed four times with DI water, after which they were treated with HCl, hot HF (20 min) to remove silicates from wind-blown dust, HCl again (to remove the colloid by-products of HF digestion), hot KOH to disaggregate clumps, two ethanol rinses to replace water and finally tertiary butyl alcohol, which was evaporated and replaced with 1995 cs silicone oil. To calculate pollen and spore concentrations (reported as number of grains/trap, because trap liquid volume varied and volumes were not preserved), a known quantity of reference material (either Lycopodium spores from batch #483316 or polystyrene microspheres) was added to each sample (Stockmarr 1971). At least 400 pollen grains and spores were counted from each trap, using a Zeiss Axio Imager at 400× magnification.

In modern systems, bison selectively reduce grasses, increasing the relative abundance of forbs (particularly Ambrosia spp. in this system) via the reduction in competition for water and light (Fahnestock & Knapp 1994). Given this, we tested whether there was a significant impact of grazing on the pollen abundances of these two taxa. Pollen from both Poaceae and Ambrosia has morphologies that do not permit identification beyond the family and genus levels, respectively; Ambrosia abundances therefore include both annual and perennial species, the latter of which includes Ambrosia psilostachya, which is an abundant forb at Konza prairie in the presence of grazing (Fahnestock & Knapp 1994).

Bison Location and Grazing Intensity

Annual records of bison biomass are available from yearly fall round-ups, when individuals are vaccinated, tagged, weighed and culled as necessary to maintain a sustainable herd size. Bison location data consisted of GPS collars placed on a subset of individuals, specifically matriarchal females, which recorded the position of the matriarchs at regular time intervals. Bison social structures change seasonally, but herds are composed primarily of cow–calf groups, small groups of bachelors or large aggregates during the breeding season. Collared bison at Konza are chosen to best reflect the distribution of the majority of the herd throughout the year. GPS collar data were the basis for the calculations of megaherbivore biomass and grazing intensity around each Tauber trap. Position data were collected at 30-min, 1-h and 2-h intervals throughout the study and were normalized for this analysis to a consistent time interval. We calculated the total number of GPS hits within concentric rings around each trap (25, 50, 100, 200, 300, 400 and 500 m). Six bison matriarchs were collared in 2008–2009 out of a total herd of 387 bison (adults and juveniles), and 11 matriarchs were collared in 2009–2010 out of a herd size of 349 bison (Table 1). From this information and annual data on herd size and total herd mass (Table 1), we calculated grazing intensity near each trap:

display math

where Ii is grazing intensity (kg m−1 year−1) at trap i, NCij is the number of collared bison in the herd at trap i (i.e. the number of GPS hits reported at a trap over the course of a year within radius j), NH is the number of bison in the herd, MH is the mass of the herd in kg, and A is the total circular source area of trap i at radius j. See Table 1 for the values of NC, NH and MH for 2009 and 2010. This formulation assumes that the number of bison and total herd biomass is fully and evenly distributed among the collared matriarchs, such that each GPS ‘hit’ for a collared matriarch reflects a fixed proportion of the overall herd biomass. Traps near the enclosure boundary may be collecting pollen and spores from both inside and outside the bison enclosure. Regardless, A represents the full circle area, including source regions both inside and outside the enclosure, regardless of trap position.

Table 1. Annual bison data
Year# CollarsHerd sizeHerd massAverage kg bison−1
20096387135 746 kg350.77 kg
201011349118 209 kg338.71 kg

Because of the limited size of the data set, we pooled all trap data for 2008–2009 (hereafter referred to by collection year, ‘2009’) and 2009–2010 (‘2010’) for all analyses, except for the three traps in the cattle-grazed watersheds from 2010. This pooling assumes that Sporormiella abundances in the traps were not significantly affected by interannual variations in environmental factors such as aridity. Moreover, using annual data for herd biomass, we do not account for seasonal variability in mean bison weight, herd size (due to the addition of calves or the culling of adults midseason) or bison land use preference. Violations of these assumptions are a source of unexplained variance in the statistical models linking Sporormiella abundances to grazing intensity.

Statistical Analyses

For these analyses, the response variable was Sporormiella abundance, represented as either absolute concentrations (# of spores/trap) or relative abundance (% or natural-log-%, as indicated). We report results for several predictor variables (representing alternate measures of megafaunal presence and grazing intensity) and statistical analyses, including (i) a Kruskal–Wallis test for a significant relationship between treatment type (i.e. whether a trap was in a bison, ungrazed or cattle watershed) and Sporormiella abundance, (ii) ordinary least-squares regression (OLS) in which the predictor variable was the area of the circle inside the bison enclosure (calculated using GIS and with separate analyses for each radius) and (iii) OLS in which the predictor variable was grazing intensity, as calculated above. We also tested OLS regression models that included elevation as a predictor variable, as well as a model where elevation was the sole predictor, in order to determine whether the effect of landscape position on spore transport was able to explain any of the observed patterns (they were not). The coefficient of determination (R2), Spearman's rank correlation rho and anova r values are reported as measures of goodness-of-fit.

Tests were performed for each circle radius, the better to assess the source area of Sporormiella. For OLS linear regression, all percentage data were converted to natural log percentages, so that the distribution would approximate the normal distribution. Logarithmic and square root transformations of percentage data are common in the analysis of compositional data (including pollen analysis) to downweight the signal from abundant types (Aitchison 2003). Response variables were tested for spatial autocorrelation using a Moran's I test (Im; Moran 1950), both on the original data and on the residuals after OLS regression. Receiver operating characteristic (ROC) analysis was conducted on the treatment data (i.e. whether traps were in grazed vs. ungrazed areas) to detect an optimal Sporormiella% threshold for the prediction of the presence of bison, selected to maximize the area under the curve (AUC). All analyses were done using R (R Core Development Team 2011), except for the ROC analysis, which was done using Sigmaplot 12.2 (Systat Software, Inc., San Jose California USA. All data and R code used in this study are available in the Supporting Information.

Results

Sporormiella abundances were significantly higher within the bison enclosure for both (untransformed) relative abundances (Kruskal–Wallis χ2 = 23.27, d.f. = 2, = < 0.001) and absolute concentrations (Kruskal–Wallis χ2 = 24.87, d.f. = 2, = < 0.001) (Fig. 2). The mean per cent abundance of Sporormiella in bison-grazed locations was 6% vs. 1.2% in the ungrazed locations and 0.7% for the cattle-grazed traps. ROC analysis indicated that 2.8% was the optimal threshold for discriminating between traps in ungrazed vs. bison-grazed locations (Fig. 3), with a strong ability to discriminate between ungrazed and grazed traps (AUC 96.8%, = < 0.0001). This threshold equally balanced test sensitivity (88.89%) and specificity (94.74%), that is, neither type I nor type II errors were prioritized (Wahl 2004). The three traps located inside cattle-grazed pastures appear to have had variable spore abundances (Fig. 2), although the sample set is too small for formal analysis and implications are difficult to assess given that data on cattle stocking densities were not available. These traps were excluded from further tests.

Figure 2.

Box plots showing the absolute concentrations (a) and relative (per cent) abundance (b) of Sporormiella from traps in bison, cattle and ungrazed treatments. The whiskers represent the upper and lower range of Sporormiella values. The bold line within each box represents the median value. The open circles represent outliers, defined as values less than Q1 – 1.5*IQR or greater than Q3 + 1.5*IQR, where Q1 is the lower quartile and Q3 is the upper quartile and IQR = Q3- Q1. For both relative and absolute concentrations, mean Sporormiella values are significantly higher in bison-grazed than in the 3 cattle-grazed or 12 ungrazed plots.

Figure 3.

(a) Receiver operator characteristic (ROC) analysis of Sporormiella abundance as a predictor of whether a trap is in a bison-grazed location. (a) A ROC plot evaluates the trade-off between high sensitivity (the true positive fraction; traps inside the bison enclosure are correctly identified as inside the enclosure) and low specificity (the false-positive fraction; traps outside the bison enclosure are incorrectly identified as inside the enclosure) for alternate discriminant thresholds applied to the Sporormiella percentages. The area under the ROC curve (a measure of the discriminant skill) is 0.97. A threshold of 2.8% maximized the area under curve and equally weights sensitivity and specificity. (b) Data used in the ROC analysis, which include bison-grazed (black circles) and ungrazed (open white circles) traps but not traps in cattle-grazed watersheds. The ungrazed sample with the highest Sporormiella value (4.7%) is from trap 743, which is located just outside a cattle-grazed watershed.

Total grazed area around the Tauber traps was a significant predictor of Sporormiella abundances for all radii (all < 0.01), and the variance explained ranged from 19.2% to 26.3% (Sporormiella concentrations) and from 34.2% to 45.6% (log Sporormiella%) (Fig. 4 and Table 2). The strongest relationship between grazed area and Sporormiella abundances occurred at the smallest circle radius tested, 25 m, for both concentration (R2 = 0.26, F1,26 = 12.12, = < 0.001) and log per cent abundance (R2 = 0.46, F1,26 = 28.94, = < 0.001) (Fig. 4, Table 2). The Spearman's rank correlation between grazed area and Sporormiella abundance was also significant, both for log percentages (rho = 0.81, = < 0.001) and concentration (rho = 0.73, = < 0.001). For both per cent and abundance data, the Spearman's rank correlation coefficient declined with increasing distance from the trap (Fig. 4).

Table 2. Sporormiella source area
Sporormiella concentration
Circle radius (m)R2Anova F1,26 P-valueSpearman's rhoP-value
250.26312.1240.0010.727< 0.001
500.24611.1050.0020.715< 0.001
1000.2109.0640.0050.693< 0.001
2000.1928.1010.0070.715< 0.001
3000.1988.4150.0060.711< 0.001
4000.2078.8520.0050.706< 0.001
5000.2159.2860.0040.704< 0.001
Log Sporormiella %
Circle radius (m)R2Anova F1,26P-valueSpearman's rhoP-value
250.46028.295< 0.0010.810< 0.001
500.45228.016< 0.0010.761< 0.001
1000.39722.377< 0.0010.734< 0.001
2000.35818.993< 0.0010.736< 0.001
3000.36019.200< 0.0010.739< 0.001
4000.36019.151< 0.0010.726< 0.001
5000.34317.724< 0.0010.696< 0.001
Figure 4.

Regression analysis of Sporormiella relative and absolute abundances vs. the amount of bison-grazed area within circles of increasing radii centred on each trap (Fig. 1). Each radius is indicated by a unique colour and shape, with increasing radii represented by the progression of colours through the colour spectrum. Both natural-log (%) abundance (a) and concentrations (c) of Sporormiella significantly and positively correlate to grazed area, with the strongest correlation at the circle radii closest to the trap (b and d, respectively). Relative Sporormiella abundance correlates more strongly than absolute abundances to grazing area. The wide range of scales compresses the values for the closest trap distances on the x-axis. However, the number of zero values makes plotting distance on a log scale unfeasible, as the slope of the regressions is sensitive to the choice of values added to zero. The changing nature of the relationship between area and spore abundance at increasing radii is best represented at a linear scale.

We found a significant positive relationship between bison grazing intensity (kg m−2 year−1) and spore abundance (Fig. 5, Table 3), but the spatial scaling of that pattern differed between per cent and concentration data. Grazing intensity showed the strongest relationship to spore concentrations at 25 m (Fig. 5c,d), which was the closest ring distance to the trap (R2 = 0.54, F1,26 = 40.83, = < 0.001); this result was similar to the pattern observed for grazed area (Fig. 4). When log Sporormiella percentages were the response variable, the strength of the relationship did not vary with distance (Fig. 5a,b), with the strongest relationship observed at 400 m (R2 =0.38, F1,26 = 20.63, = < 0.001). Elevation was not a significant predictor of spore abundances in a multiple regression model including grazing intensity at 25 m (for grazing intensity, β = 63.043, = < 0.0001, while for elevation, β = −3.233, = 0.907) or in an OLS model of elevation as a predictor alone (R2 = 0.03, = 0.311).

Table 3. Sporormiella and grazing intensity
Sporormiella concentration
Circle radius (m)R2Anova F1,26P-valueSpearman's rhoP-value
250.54640.835< 0.0010.805< 0.001
500.45428.276< 0.0010.761< 0.001
1000.41424.063< 0.0010.811< 0.001
2000.24711.1600.0020.666< 0.001
3000.26612.2910.0010.687< 0.001
4000.23810.5970.0030.695< 0.001
5000.2058.4850.0060.668< 0.001
Log Sporormiella %
Circle radius (m)R2Anova F1,26P-valueSpearman's rhoP-value
250.25711.7820.0020.768< 0.001
500.28713.465< 0.0010.704< 0.001
1000.37720.588< 0.0010.788< 0.001
2000.35418.612< 0.0010.788< 0.001
3000.36219.262< 0.0010.694< 0.001
4000.37820.635< 0.0010.717< 0.001
5000.33816.874< 0.0010.677< 0.001
Figure 5.

Regression analysis of Sporormiella abundance vs. grazing intensity (kg/m2/year) assessed at increasing radii from each trap. As in Fig. 4, each radius is indicated by a unique colour and shape. Both natural-log(%) abundance (5a) and concentrations (5c) of Sporormiella positively correlate to grazing intensity. For Sporormiella relative abundances (5b), this relationship does not vary significantly across radii, while Sporormiella concentrations (5d) show the strongest significant relationship between Sporormiella and grazing intensity closest to the traps (25 m).

The regression coefficients relating log abundance and concentration to grazed area exhibited a decline in the slope of the relationship with increasing values of grazed area (Fig. 5a, c). When grazed area was converted to biomass per unit area, the log spore/biomass relationship (F1,237 = 113.2, < 0.001) did not vary by ring width (F6,0.26, > 0.05; Fig. 5a). The pattern of decreasing slopes with grazed area may indicate dispersal limitation of Sporormiella spores, such that the relationship between spore production from dung and deposition at a point location degrades with increasing radii of dung deposition. When biomass per unit area was considered, the distance dependence was removed, because the biomass value was standardized across the entire ring area.

The abundances of the spores in the original data across all traps also showed low spatial autocorrelation (spore concentrations, Im = 0.156, = 0.017; per cent abundance, Im = 0.193, = < 0.001). The residuals of the regression models of spore per cent abundances and concentrations as predictors of grazing intensity showed low and non-significant spatial autocorrelation at all circle radii (e.g. residuals at 25: concentration model, Im = 0.089, = 0.08; per cent abundance model, Im = 0.024, = 0.256).

Abundances of Ambrosia pollen were significantly higher in traps located in grazed than in ungrazed watersheds, for both concentration (Kruskal–Wallis χ2 = 8.29, = 0.004) and untransformed percentage data (Kruskal–Wallis χ2 = 9.42, d.f. = 1, = 0.002) (Fig. 6). Conversely, Poaceae per cent abundances were significantly lower in grazed than in ungrazed sites (Kruskal–Wallis χ2 = 5.18, d.f. = 1, = 0.023), but concentrations were not significantly different (Kruskal–Wallis χ2 = 0.73, d.f. = 1, = 0.39) (Fig. 6).

Figure 6.

Box plots of Poaceae concentrations (a) and Ambrosia concentrations (b) and relative abundances (per cent) of Poaceae (c) and Ambrosia (d) by treatment. Mean Ambrosia concentrations are significantly higher in bison-grazed than in ungrazed traps, while Poaceae abundances are higher in ungrazed traps. Mean Ambrosia per cent abundances are significantly higher in bison-grazed than in the ungrazed traps, while Poaceae per cent abundances are lower, but not significantly so. Figure format with respect to use of whiskers and open circles follows usage in Fig. 2.

Discussion

Understanding the Sporormiella Proxy

This study reinforces prior findings that the abundance of Sporormiella spores in depositional environments reflects the presence and number of large mammalian grazers and browsers (Graf & Chmura 2006; Raper & Bush 2009; Parker & Williams 2012; Etienne et al. 2013). This relationship is robust even in a relatively dry system such as the tallgrass prairie (mean annual precipitation at Konza is ~850 mm). Moreover, the detailed data available at Konza on bison biomass and landscape preference allow this study to move beyond the presence/absence correspondences reported in prior papers and demonstrate both a positive correlation between megaherbivore grazing intensity and Sporormiella abundances and a palynological signal of the effect of grazing on grassland composition.

The strong differences in Sporormiella abundances between traps inside and outside the bison enclosure (Fig. 2) clearly show that at Konza, the contribution of megaherbivores to Sporormiella values over-rides any signal associated with the activity of smaller or more transient herbivores – an important point, given that Sporormiella has many modern vertebrate hosts (Ahmed & Cain 1972). Other common mammalian herbivores at Konza include white-tailed deer (Odocoileus virginianus), eight species of rodents and the eastern cottontail (Sylvilagus floridanus) (McMillan et al. 1997). Given that Konza is not a single herbivore system (though it is a single keystone herbivore system), the robustness of Sporormiella as a predictor of bison presence alleviates concerns that late Quaternary Sporormiella fluctuations were driven by changes in the abundances of smaller mammalian herbivores, such as lagomorphs (Feranec et al. 2011). Rather, small herbivores appear to set a background level of Sporormiella, beyond which a megafaunal signal can be detected. The threshold value of 2.8% indicated by the ROC analysis of the trap data is consistent with the 2% threshold proposed by Davis (1987) for identifying megafaunal extirpations from Sporormiella in late Quaternary sediments (note that the 2.8% threshold reported here was based on the sum of all pollen and spores, while Sporormiella is typically reported as a percentage of the upland pollen sum alone in palaeorecords). We also processed bison dung for this study; Sporormiella spore percentages in the traps were lower than the abundances in the bison dung (16%), as well as the abundance of Sporormiella in mammoth coprolites (16%) reported by Davis (1987).

The existence of an association between Sporormiella abundances and megaherbivory is robust to the choice of response variable (both absolute concentrations and relative abundances of Sporormiella) and to the choice of predictor variable (grazed vs. ungrazed treatments, area grazed around traps and grazing intensity) and does not appear to be an artefact of elevation or spatial autocorrelation. However, the form and strength of this relationship vary depending on whether relative (percentage) or absolute (concentration) abundances are used. The stronger correlation of concentration data to bison grazing intensity (Table 3) suggests an effect of grazing on vegetation structure that may be influencing pollen accumulation rates for some plant taxa and hence partially confounding the relative abundance data. Because pollen percentages are affected by the pollen accumulation rates of other taxa while absolute concentrations are not, we hypothesize that the different relationships observed for the percentage and concentration data may be because (i) the Sporormiella percentages are confounded by the influence of grazing pressure on grassland composition and hence the pollen accumulation rates of other taxa and (ii) percentage data may be more heavily influenced by vegetation differences due to topographic heterogeneity.

Our work confirms prior inferences that Sporormiella has a short dispersal distance (Raper & Bush 2009; Parker & Williams 2012). The presumed primary modes of transport in general are short-distance wind transport, slopewash and saltation; spores that enter Tauber traps are likely to be via wind, not water (Tauber 1974), although there may be a slopewash component during heavy rains (see Fig. S1 for a photograph showing the Tauber trap opening relative to the ground surface). Our results (Fig. 4) suggest that the relevant source radius (Sugita 1993) of Sporormiella is on the order of 25 m or less, which indicates that they are a very local indicator of the presence of megaherbivores in palaeorecords. Our study does not directly test the source area of spores due to slopewash or account for the fact that there is evidence that megaherbivores may be directly defecating into lakes. For example, Raper & Bush (2009) found that spore abundances were higher closer to the shoreline than in the deeper portion of lakes, presumably due to the fact that unlike wind-transported pollen, spores are either being deposited directly into the lake via defecation or are washed in from the surrounding shoreline. Given that pollen transport into lakes is primarily via wind and can be transported very long distances (> 100 km), our study does suggest that a long-distance wind-dispersed component of Sporormiella is probably negligible, due to the fact that spores are released so close to the ground and may never leave the boundary layer of relatively still air created by vegetation (although this may not be the case in grazing lawns, where grass is cropped close to the ground surface).

The Tauber traps in this study and the lakes and mires sampled in fossil pollen analysis both primarily receive aerially transported pollen, but there are important differences in their depositional contexts. In lakes, pollen is mixed within the water column over seasons or even years before finally being incorporated into the sediment record (Sugita 1993). Within lakes, the pollen source area includes both local and regional vegetation, and the relevant source area increases with lake size due to the increased relative contribution of regional pollen (Sugita 1993; Davis 2000). Conversely, Tauber traps are non-volumetric samplers with small (typically 5 cm) openings and may be more susceptible to variations in wind speeds and surface turbulence and therefore pollen grain size and fall speed, which vary among taxa (Levetin, Rogers & Hall 2000). Because trap openings are very small relative to lakes, the source area of pollen is much more local for traps than for lakes (Tauber 1974; Hicks & Hyvarinen 1986; Sugita 1993). One standard assumption in the use of Tauber traps is that trap aerodynamics minimize collection biases (Tauber 1974; Hicks & Hyvarinen 1986; Levetin, Rogers & Hall 2000). In this study, this assumption may be violated because bison grazing strongly affects vegetation structure (Fig. S1), which may affect the transport and deposition of pollen and spores to the trap. In short, Tauber traps tend to collect both pollen and Sporormiella locally, as opposed to lakes, which have a mixture of local and regional pollen sources but presumably a local Sporormiella source area. Therefore, the statistical relationships reported here should not be applied to explicitly convert Sporormiella abundances retrieved from lake sediments to estimates of herbivore grazing intensity or biomass. Nevertheless, the quantitative link between increased Sporormiella abundances and bison grazing intensity reported here validates the use of the Sporormiella spores in pollen analysis as a proxy for megaherbivory.

Implications for Megaherbivory and Climatic Drivers of Holocene Vegetation Dynamics

The use of Sporormiella preserved in sediments will allow for an in situ investigation of the role of megaherbivores in determining past vegetation dynamics. In particular, independent metrics of herbivore biomass and vegetation composition from the same depositional environment will enable enquiries into the significance of megaherbivory as a top-down regulator of vegetation dynamics during the Quaternary (Owen-Smith 1987; Gill et al. 2009, 2012; Johnson 2009). The role of bison on the Great Plains during the Holocene has been particularly difficult to determine against a dynamic background of climate variability (Craine & McLauchlan 2004; Grimm, Donovan & Brown 2011). As one of the few megafaunal survivors of the late Quaternary extinctions in North America (Koch & Barnosky 2006), bison provide a rare opportunity to continuously study the interactions between vegetation, megaherbivores and climate variability since the last deglaciation. Several well-dated, high-resolution Great Plains lake sediment records show 100–160 year oscillations of drought variability during the early to middle Holocene (9.8–2.8 ka BP); these are recorded in the pollen record by alternating high Poaceae abundances during wet intervals and high Ambrosia abundances during dry intervals (Clark et al. 2002; Nelson et al. 2004; Grimm, Donovan & Brown 2011). When interpreted climatically, this pattern has been called the ‘Ambrosia paradox’ (Grimm 2001) because Ambrosia and other forbs are less able to compete with grasses for moisture (Fahnestock & Knapp 1994) and so would be expected to decline during dry periods. Grimm, Donovan & Brown (2011) hypothesized that increased moisture variability during drier intervals may have favoured the expansion of Ambrosia by creating more disturbed habitat and exposed shoreline. Craine & McLauchlan (2004), however, proposed an alternative hypothesis that the observed Ambrosia–Poaceae cycles were instead driven by population dynamics in bison, during which periods of intense grazing suppressed Poaceae, favoured Ambrosia and reduced available fuels for fire. Our finding that the effects of megaherbivory on Poaceae and Ambrosia can be detected in pollen records removes one possible objection to the Craine & McLauchlan (2004) hypothesis while leaving the main question (i.e. the relative importance of abiotic and biotic controls on Holocene grassland dynamics) still unresolved.

How can strong evidence that drought variations were closely linked to past grassland and fire dynamics be reconciled with equally strong evidence that in contemporary systems bison grazing is a critical determinant of grassland community composition and function? Evidence of past cycles in moisture availability, vegetation composition and fire dynamics is not mutually exclusive with the modern evidence that bison grazing is a critical determinant of grassland community composition and function. One possible way to reconcile these lines of evidence is to consider that climate variability is regulating both bison populations and behaviour and grassland composition during the Holocene, but that bison act as a key mediator of the effect of climate on vegetation during dryer conditions. In this hypothesis, drought would act to alter the spatial heterogeneity of the effect of bison on grassland composition, while bison act to enhance the relative abundance of Ambrosia during dry conditions, by both (i) reducing below-ground competition for water through selective removal of grasses and (ii) increasing the production of wallows and other disturbed patches, which provide optimal Ambrosia habitat during drought conditions (McMillan, Pfeiffer & Kaufman 2011). Periods of water stress are common in the prairie and can enhance the effects of herbivory on vegetation (McNaughton 1983; Frank & McNaughton 1992). Grazing, wallowing and burning facilitate landscape heterogeneity by creating patches at different levels of disturbance (Collins & Calabrese 2012). During dry periods in the Holocene, lake levels across the Great Plains were reduced, and many lakes dried up when aridity was most severe (Digerfeldt, Almendinger & Bjorck 1993; Valero-Garcés & Kelts 1995; Haskell, Engstrom & Fritz 1996; Laird et al. 1996). Hence, drought may have altered the spatial scale of grassland patch dynamics, bringing bison – and therefore the patchy landscapes associated with bison activity – closer to lakes that remained potable.

Conclusions

Our results (i) demonstrate the first quantitative link between Sporormiella abundances and megaherbivore biomass, for both percentage and concentration data, (ii) support a local (< 100 m) source area for Sporormiella, (iii) show that high Sporormiella abundances can be linked to the sustained presence of megaherbivores and that this signal is detectable above a background level contributed by mesofauna, iv) support the previously reported 2% threshold as a cut-off for megafaunal presence in pollen records and v) establish that some of the well-documented effects of bison herbivory on vegetation can be detected in the palynological record.

This study refines the use of the Sporormiella proxy to test the relationship between megaherbivores and vegetation, an emerging subdiscipline of palaeoecology. A growing number of studies, including this one, suggest that biotic interactions like herbivory can leave a measurable signature in the palaeorecord. Process-based models of the drivers of Quaternary vegetation dynamics should be expanded to include megaherbivores, as well as interactions between megafauna and climate. The grasslands of the Great Plains were one of the few habitats in North America to sustain large populations of megaherbivores following the end-Pleistocene extinctions, and therefore, palaeorecords from this region represent a unique opportunity to test hypotheses about the interactions between climate variability and keystone herbivores on vegetation dynamics at centennial to millennial time-scales.

Acknowledgements

This work was supported by NSF DEB-0716471, the NSF LTER Program at Konza, the University of Wisconsin Geography's Department Whitbeck Fellowship and the Bryson Professorship of Climate, People, and the Environment. We thank the staff at Konza Prairie for logistical support and access to data; Eli Martinson, Nancy Parker and many KSU undergraduates for trap assistance; Guy Robinson for Sporormiella reference slides; Jim Burt, Joe Mason, Steve Jackson, Sara Hotchkiss, David Mladenoff, Warren Porter and members of the Williams Lab for valuable commentary; and Mark Bush and an anonymous reviewer, whose suggestions greatly improved the manuscript.

Ancillary