There has been much debate on the impact of large herbivores on biodiversity, especially given that large mammals are becoming locally extinct in many places.
The use of evolutionary information on community structure has typically been limited to evaluating assembly processes, for example, competition or habitat filtering, whereas a lack of long-term experiments has precluded the test of predictions considering more complex biotic interactions.
Reconstructing the complete phylogeny of the trees and shrubs of the Kruger National Park from DNA data, we tested for phylogenetic signal in antiherbivory traits and compared the phylogenetic structure of communities under various degrees of herbivore pressure using experimental plots spanning several decades.
We show that all antiherbivory traits examined demonstrated weak but significant phylogenetic signal, and that exclusion of large herbivores results in impoverished species diversity in restructured communities. Surprisingly, we also show that reduction in species diversity coupled with community reorganization does not necessarily result in a decrease in phylogenetic diversity, and that community responses to herbivore exclusion depend on initial structure.
Synthesis. Extinction of large mammal herbivores will have cascading effects on plant diversity; however, impacts on plant community structure are contingent on initial conditions. This research has implications for best practice when managing large herbivores and natural habitats.
Phylogenetic information is increasingly being used to study patterns and processes of community assemblages (Webb et al. 2002; Cavender-Bares et al. 2004, 2009; Morlon et al. 2011). Here, we focus on the subtropical woodland biome of the Kruger National Park (KNP) in South Africa (Schmidt, Lotter & McCleland 2007). This biome is commonly referred to as ‘bushveld’ in South Africa, but it is widely distributed across the continent, and is home to the largest terrestrial mammals, including elephants, rhinos, giraffes and many species of antelopes (Owen-Smith & Ogutu 2003). The vegetation varies from dense thicket, savanna woodlands to forest with tall trees and closed canopy (Schmidt, Lotter & McCleland 2007). The dynamics of these plant communities are dictated by various disturbances (Milchunas, Sala & Lauenroth 1988; Du Toit 2003), including periodical events such as fire (Govender, Trollope & Van Wilgen 2006) and more or less continuous pressures from large herbivores (Carson & Root 2000; Van Langevelde et al. 2003). Recent studies have shown the important role of fire in structuring the savanna ecosystems (e.g. Collins et al. 1998; Govender, Trollope & Van Wilgen 2006) and emphasised the effects of herbivory on plant diversity (e.g. Collins et al. 1998; Proulx & Mazumder 1998; Howe, Brown & Zorn-Arnold 2002; Adler et al. 2005; Kohyani et al. 2008). Although large herbivores have been shown to reduce the three-dimensional structure of vegetation in the KNP (Asner et al. 2009), several studies suggest that herbivores may favour the diversity of woody plants (e.g. Sankaran et al. 2005) whilst others suggest negative impacts (e.g. Bond & Keeley 2005; Levick & Rogers 2008). However, the impacts of herbivory on phylogenetic structure of plant assemblages are less clear (Cavender-Bares et al. 2009). Recent works have shown a link between phylogenetic diversity and ecosystem function in plant communities (Cadotte et al. 2009; Flynn et al. 2011), and there is a growing body of literature, which suggests that community phylogenetic structure might influence community invasibility (e.g. Lessard et al. 2009). Understanding how phylogenetic community structure will be altered with the removal of large herbivores will therefore be critical for predicting ecosystem processes and community dynamics (Cavender-Bares et al. 2009).
Here, we employed a phylogenetic framework to characterize the community of woody plant species in the KNP and evaluated the impacts of the removal of large herbivores on community composition and structure. Analysing how large herbivores alter the phylogenetic structure of plant communities may provide a better understanding of ecosystem processes and community dynamics. Different phylogenetic patterns are expected under various regimes of herbivory and plant defences (Cavender-Bares et al. 2009; Fig. 1). First, if most herbivores are generalist (i.e. a broad range of plant species is grazed), plants with generalist defences, such as spines, will tend to dominate the community. If these generalist defences are evolutionarily conserved, such that closely related plant species share similar defences, the community will then tend to be phylogenetically clustered – composed of more closely related species. However, when defence traits are convergent, the community may demonstrate greater phylogenetic overdispersion, such that community members are less closely related. Second, when plants face high specialist herbivory pressures, only plant species falling within herbivore dietary preferences are targeted. Under such a scenario, rare and non-preferred species could escape herbivory and ultimately increase local plant diversity. Under multiple specialist herbivore pressure, when defence traits are both matched closely to a specific herbivore and tightly conserved within plant clades, the community will be phylogenetically overdispersed, but less structured when traits are convergent (Cavender-Bares et al. 2009).
The KNP is renowned for its large game animals. There are 148 mammal species, of which 30 are large herbivores (Du Toit 2003; see Table S1 in Supporting Information), defined here as herbivorous mammals weighing over 5 kg (as opposed to megaherbivores usually defined as weighing over 1000 kg; Owen-Smith 1988). Indeed large herbivores are recognized as ecosystem engineers (Waldram, Bond & Stock 2008) or keystone ecosystem species (Owen-Smith 1988), although abiotic factors, including climate, soil and disturbances such as fire, are also important in structuring savanna ecosystems. The KNP is one of the largest nature reserves in the world, and various habitats are found within its 20 000 km2. The flora of the KNP comprises 1974 plant species including 458 species of trees and shrubs. We used DNA sequencing to generate a phylogenetic tree of all woody species and characterized their community structure and diversity across a north–south transect spanning c. 350 km. We then evaluated the impacts of herbivory by contrasting changes in woody plant communities within long-term ecological plots, from which large herbivores have been excluded for several decades (‘exclosures’), to the immediately surrounding park. Our study is limited by the existing set up of exclosures in the park, but it represents the first phylogenetic analysis of the effect of large herbivores on plant communities at the landscape scale.
Materials and methods
The KNP is situated in the north-eastern part of South Africa between 22°25′ and 25°32′S and 30°50′ and 32°E. It is part of the ‘Greater Maputaland–Pondoland–Albany’ biodiversity hotspot (Perera, Ratnayake-Perera & Procheş 2011). Rainfall varies from 440 mm in the north to 740 mm in the south (Venter 1990). Mean annual temperature is around 21─23 °C, but in summer temperatures often exceed 38 °C, and frost can occur sporadically during winter. Exclosures have been established in the park for between 8 and 43 years – where large herbivores are partly or fully excluded. All large herbivores are fully excluded from three exclosures (Table S2): Hlangwine (220 ha; 38 years old), Nkuhlu 1 (70 ha; 8 years old) and Nwashitsumbe (302 ha; 43 years old). In two partial exclosures, Nkuhlu 2 (44 ha) and Letaba (129 ha), megaherbivores such as elephants and giraffes have been excluded for 8 years, but other large herbivores can still gain access (the fence goes from 1.50 m upwards).
Phylogeny of the KNP Flora
We reconstructed the phylogeny of 448 species of trees and shrubs plus Amborella used as out-group to root the tree (APG III 2009); these represent 246 genera, 71 families and 31 orders (sensu APG III 2009; Table S3; Fig. S1). Species identification was verified using DNA barcoding. We sequenced two DNA regions: a portion of matK (942 base pairs including gaps) and subunit ‘a’ of rbcL (552 base pairs including gaps). These regions have been identified as suitable ‘DNA barcodes’ for land plants but also for phylogeny reconstruction (CBOL Plant Working Group 2009). DNA extraction from leaves, polymerase chain reactions and sequencing were conducted using standard protocols (Hajibabaei et al. 2005; Ivanova, Fazekas & Hebert 2008). All GenBank/EBI accession numbers for gene sequences (accession numbers JF265241–JF265667 for rbcLa and JF270599–JF271008 for matK) and voucher information (including photographic images) are available online from the African Centre for DNA Barcoding (www.acdb.co.za) and listed in (Table S3). Phylogeny reconstruction based on combined matK + rbcLa data was performed with maximum likelihood (ML) (Felsenstein 1973) using RAxML-HPC2 vs. 7.2.6 (Stamatakis, Hoover & Rougemont 2008) on the CIPRES cluster (Miller et al. 2009).
Divergence times were estimated using a Bayesian MCMC approach implemented in beast (v. 1.4.8; Drummond & Rambaut 2007), which allows simultaneous estimation of the topology, substitution rates and node ages (Drummond & Rambaut 2007). We selected GTR + I + Γ as the best model of sequence evolution for each partition based on the Akaike information criterion (AIC) evaluated using modeltest (V 2.3; Nylander 2004). A speciation model following a Yule process was selected as the tree prior, with an uncorrelated lognormal (UCLN) model for rate variation among branches. We used the following secondary calibration points with a normal prior distribution based on Bell, Soltis & Soltis (2010): Oleaceae crown node (41 Ma, SD 6 Ma), Moraceae crown node (31 Ma, SD 4 Ma), Rhamnaceae crown node, Malvaceae crown node (39 Ma, SD 4 Ma), Vitaceae crown node (43 Ma, SD 9 Ma), angiosperm crown (149 Ma, SD 3 Ma). Monte Carlo Markov Chains were run for 100 million generations with trees samples every 1000 generations. Convergence was checked using tracer v. 1.5 and of the resulting 100 001 trees, we removed 15 000 as burnin and combined the remaining 85 001 trees using treeAnnotator v.1.7.1 (Rambaut & Drummond 2008).
We surveyed 110 2500 m2 plots (50 × 50 m) distributed throughout the KNP following a north–south transect (Fig. S2), and sampling from each of 15 ‘ecozones’, defined by vegetation and soil types (Venter 1990). Plots were spread within each ecozone but accessible by gravel roads; to avoid edge and fire effects, plots were situated at least 300 m from the nearest track in unburnt areas. In each plot, we recorded all species of trees and shrubs and the number of individuals per species (abundance). These KNP plots represent woody plant communities where herbivory is unrestricted. Next, we established 15 2500 m2 plots in each of the five exclosures (except in ‘Nkuhlu 2’ where 13 plots were established; Table S2) and sampled plants as for the 110 KNP plots. The KNP is a heterogeneous environment with regard to geology, geomorphology, climate, vegetation and soil (Du Toit 2003); the exclosures were designed to capture this heterogeneity. It was of course not practical for us to establish new exclosures from which large herbivores would have been excluded over several decades; we have been therefore restricted to contrasting communities across these pre-existing exclosures. In total, we sampled 183 plots, representing richness and abundance estimates for 216 species (the subset of KNP woody plant diversity found in our plots) across 457 500 m2.
Antiherbivore Defence Traits
We evaluated various physical and mechanical plant ‘defence traits’. We collated data on spinescence (presence/absence) and plant maximum height from literature (Schmidt, Lotter & McCleland 2007) as our measures of plant physical defences (see details in Data S1). We used wood density to quantify plant resistance to physical damages (Chave et al. 2009) such as breaking of branches and trunks, bark stripping and uprooting, especially from elephants. We measured specific leaf area (Wright et al. 2004) as a proxy for leaf economic spectrum, as it correlates with leaf nutrient content (Reich, Walters & Ellsworth 1997) and herbivores preferentially forage for high-quality leaves (Grant & Scholes 2006; Levick & Rogers 2008). We did not consider chemical defences here because of practical constraints; in any case woody plants, especially from nutrient-poor environments, appear to invest more in physical and mechanical defences against large herbivores than in chemical ones (Christoph & David 1997). Detailed methodology is provided in the Supporting information.
First, we evaluated the degree of phylogenetic signal in plant herbivore defence traits. Several tests have been developed to quantify phylogenetic signal, but no single test is able to account for all models of evolutionary processes (Krasnov, Poulin & Mouillot 2011). Therefore, we assessed the phylogenetic signal in continuous traits (height, specific leaf area and wood density) using two alternative phylogenetic tests: the K statistic (Blomberg, Garland & Ives 2003) implemented in the r package Picante 1.2 (Kembel et al. 2010), and Pagel's lambda implemented in the r package Geiger 1.0 (Harmon et al. 2008). In addition, we tested phylogenetic signal in the binary trait (spine presence/absence) using the D statistic of Fritz & Purvis (2010) implemented in the r package Caper (Orme et al. 2012). Both the K statistic and Pagel's lambda evaluate the signal in a trait against a Brownian motion model. The statistical significance of the K values was evaluated by comparing observed patterns of the variance of independent contrasts of the trait to a null model of shuffling taxa labels across the tips of the phylogeny. Significance of lambda was tested against a null of lambda = 0 (no signal) using the likelihood ratio test. The D statistic from Fritz & Purvis (2010) allows us to compare the observed distribution of trait values to both a random shuffle of trait value at the tips of a phylogeny and a Brownian threshold model (BM): D = 1 indicates a random trait distribution across the tips; D = 0 corresponds to a BM model; D < 0 when traits are highly conserved, and D > 0 is indicative of a phylogenetic overdispersion.
Second, we calculated three diversity metrics: the species richness (SR), the Shannon diversity index (henceforth referred to as ‘Shannon’) and the mean phylogenetic distance between taxa (MPD) in each community. Shannon differs from SR in that, in addition to depicting the number of species in each community, it also takes into account the relative abundance of each species. MPD, however, evaluates the mean branch length (mean evolutionary distance) separating all pairs of species within a community. We use MPD rather than Faith's (1992) phylogenetic diversity (PD) because it is less strongly correlated with SR.
Phylogenetic community structure was quantified using the net relatedness index (NRI) and net taxon index (NTI) (Webb et al. 2002; Webb, Ackerly & Kembel 2008) as implemented in the r package Picante 1.2 (Kembel et al. 2010). NRI describes a tree-wide pattern of dispersion of a community relative to expectations from random assemblages, whereas NTI reflects the phylogenetic structure closer to the tips. For both metrics, positive values indicate that closely related species co-occur more often than predicted by chance (phylogenetic clustering), whereas negative values indicate greater co-occurrence of more distantly related species (phylogenetic overdispersion). Here, we assessed significance of NRI and NTI comparing, respectively, the quantile of observed mean phylogenetic distance (MPD) and mean nearest taxon distance (MNTD) vs. the expected MPD and MNTD from randomly generated communities of equivalent richness. These random communities were constructed based on 10 000 simulations using two alternative regional pools: the phylogeny pool (all the tree species in the KNP) and the sample pool (only the 216 tree species found in the 110-KNP plots).
We explored the distribution of diversity metrics in several ways. We assessed how the diversity metrics (Shannon, SR, MPD) and community structure (NRI) change along a latitudinal gradient using the Pearson product–moment correlation test, with Bonferroni correction for multiple tests. Next, we evaluated the effects of large herbivores on diversity and community structure. To correct for pseudo-replication – plots within exclosures may be considered as pseudo-replicates – we used two approaches. We evaluated pattern at each site by comparing plots inside vs. outside exclosures (i.e. plots adjacent to exclosures). We defined adjacent plots as those falling within a maximum of 25 km radius of each exclosure because we found that this distance provides the best compromise between maximizing the sample size of included plots and restricting contrasts to more or less comparable communities. Statistical significance for each comparison was evaluated using a Mann–Whitney U-test and applying Bonferroni correction for multiple comparisons. We also fitted a linear mixed effect model to our data as an alternative analysis to the inside–outside comparison. For this purpose, our response variables were Shannon, SR, MPD or NRI with ‘treatment’ (exclosure vs. KNP) as explanatory variable: ‘location’ (south vs. north) and ‘site’ (defined as ecozones A-P; see Table S4) as random effects.
The phylogenetic tree reconstructed using a Bayesian approach was highly congruent with the latest phylogenetic studies of angiosperms (APG III 2009; Figs 1 and S1; data available on TreeBase ID 11232). All traits related to herbivory demonstrated significant albeit weak phylogenetic signal (Table 1).
Table 1. Tests of phylogenetic signal in plant defence. For spines, the significance of the D value was tested against simultaneously a random shuffle of traits along the tip of the phylogeny and a Brownian motion (BM) model; two P-values are therefore indicated for the trait presence/absence of spines, the first for the test against random shuffle and the second against BM model
Fritz & Purvis' D
P (random shuffle/BM)
* < 0.05; ** < 0.001; *** < 0.0001.
Specific leaf area
Across the KNP, we found a latitudinal gradient in the diversity of woody plants such that there was a trend for decreasing diversity (Shannon and SR) moving northward: Shannon (r = −0.28, P =0.012); SR (r = −0.17, P =0.32; Fig. S3); an opposite but non-significant trend was observed for MPD (r = 0.19, P =0.20; Fig. S4). To help visualize spatial variation in diversity across the KNP, we mapped interpolated values using Ordinary Kriging with a 12-cell neighbourhood (Fig. 2). The interpolated values should be interpreted cautiously (all statistics were performed using the plot-level data); however, they confirm the strong north–south gradient in the diversity of woody plants for both Shannon and SR, with highest diversity in the south and extreme north of the park and low diversity in the centre (Fig. 2). Shifts in community phylogenetic structure (NRI) broadly paralleled changes in community diversity (Fig. S4), but we observed lower NRI in the south and north of the park, and higher NRI towards the centre, indicating that communities in the centre are more phylogenetically clustered (Fig. 2). Across all plots the vast majority of NRI and NTI values were positive, indicating that co-occurring species in the KNP are generally more closely related than expected by chance (Table S4).
In a global comparison of plots within exclosures vs. plots outside exclosures, we found that Shannon and SR were consistently lower inside exclosures (Mann–Whitney; Shannon: U = 5291, P =0.0002; SR: U = 5629, P <0.001; Fig. 3), but there was no obvious trend with species evolutionary relatedness (Mann–Whitney; MPD: U = 4165, P =0.67; NRI: U = 4511, P =0.15; Fig. 3). The linear mixed effect model generated similar results with a significant difference in Shannon and SR within and outside exclosures (Shannon: P = 0.02; SR: P =0.0003; Table 2). Again, there was no obvious trend for MPD and NRI (P =0.13 and 0.89, respectively; Table 2).
Table 2. Global comparison of exclosures vs. KNP based on the linear mixed effect model
MPD, mean phylogenetic diversity; NRI, net relatedness index. * < 0.05; ** < 0.001; *** < 0.0001.
To assess further shifts in community composition, we used pairwise comparisons of plots within exclosures and their adjacent areas (Fig. S2). Confirming the global trend, we found large and significant decreases in diversity (Shannon and SR) within most exclosures, although trends for Shannon were less pronounced (Fig. 4; Table 3). The Letaba exclosure was an exception to this trend; although SR was marginally lower within this exclosure, we observed a significant increase in MPD (Table 3). Our results for the phylogenetic metric of community structure (NRI) were more complex (Fig. 5 and Table 3). In the south, no difference was found between plots inside and outside exclosures (Nkuhlu 1, U = 137 P =0.28; Nkuhlu 2, U = 144, P =0.55; Hlangwine, U = 121, P =0.13). However, in the centre of the park, we found that NRI was significantly lower within the exclosure (Letaba) than in the adjacent plots (U = 84, P =0.001), whilst in the north of the park, we found the opposite pattern, such that NRI values within the exclosure (Nwashitsumbe) were significantly higher than in adjacent plots (U = 21, P =0.0008).
Table 3. Comparisons of plots within exclosures and their adjacent areas
Values indicated are median of each metric; Shannon, Shannon diversity index; SR, Species richness; MPD, Mean phylogenetic diversity; NRI, Net relatedness index.
Exclosures differ in age (8–43 years old) and in the type of large herbivores that are excluded (full vs. partial exclosures). The limited number of exclosures (n = 5) precludes full statistical testing; nonetheless, in pairwise comparisons, we found no significant difference in diversity patterns between exclosures of the same type (i.e. full exclosures) but of different ages (8 vs. 38 years i.e. Nkuhlu 1 vs. Hlangwine; SR: U = 131, P =0.12; MPD: U = 16659.5, P =0.9; Shannon: U = 127, P =0.18; Fig. S5), or between partial vs. full exclosures, even when comparing adjacent exclosures of equivalent age at the same site (e.g. Nkuhlu 1 vs. Nkuhlu 2; SR: U = 109, P =0.69; MPD: U = 16642.5, P =0.93; Shannon: U = 98, P =1; Fig. S6). In addition, we also found no evidence for difference between exclosures of the same type (e.g. full exclosure) and similar age (38 vs. 43 years), but in different locations (i.e. Hlangwine in the south, granitic soil, relatively higher rainfall vs. Nwashitsumbe in the north, basaltic soil, lower rainfall: median NRI = 2.21 and 1.81 for Nwashitsumbe and Hlangwine respectively; U = 151, P =0.12).
There is increasing evidence that communities are generally clustered (see review in Vamosi et al. 2009), and that phylogenetic signal in biological traits is common (see review by Wiens et al. 2010). Across the KNP where herbivory is unrestricted, we found significant phylogenetic clustering of plant communities. Clustering has been attributed to a variety of mechanisms, including habitat filtering (Webb et al. 2002), disturbance (Helmus et al. 2010), facilitation (Valiente-Banuet & Verdu 2007), competition and biotic interchange (Kissling et al. 2012) or a combination thereof (Mayfield & Levine 2010). One recent theory predicts phylogenetic clustering of plant communities under heavy pressure from generalists when plant defence traits are evolutionarily conserved (Cavender-Bares et al. 2009). The majority of large herbivore browsers in the KNP are generalists, but the strength of phylogenetic conservatism in ‘defence traits’ has not previously been assessed.
Several plant attributes have been identified as herbivore defence traits. These include various physical and mechanical properties, low leaf nutrient content and various chemical compounds (Rosenthal & Kotanen 1994). Here, we used multiple approaches to evaluate phylogenetic signal in key defence traits. In aggregate, we found that most defence traits demonstrate weak but significant phylogenetic signal, perhaps indicating that they do not fit well to a simple model of Brownian motion evolution. For example, phylogenetic structure in trait distributions may be driven by one or few clades. We suggest that although the evolution of any one trait may be idiosyncratic (i.e. demonstrate evolutionary trajectories that depart from strict Brownian motion, but nonetheless still retain some imprint of phylogenetic history), phylogenetic distances between taxa might still provide a reliable indicator of species differences and similarities in plant defence strategies aggregated across multiple traits, all of which covary somewhat with phylogeny (Felsenstein 1988).
Our findings of phylogenetic clustering of plant communities, significant (albeit weak) phylogenetic signal in defence traits and evidence for strong generalist herbivory pressure imposed by large herbivores match predictions from Cavender-Bares et al. (2009) and indicate that large herbivores may have played a significant role in shaping the structure of plant communities in the KNP. However, strength of community clustering is variable across the KNP likely reflecting the variable geomorphology of African savanna (Venter 1990) and the patchy distribution of large herbivores in the park (Turner et al. 1997; Levick & Rogers 2008). Unsurprisingly, plant diversity within the KNP is also strongly spatially structured.
Although diversity, as indexed by Shannon's index and SR is highest in the north and south of the park, and lowest towards the centre of the park, there is a general latitudinal trend towards higher diversity in the south. This gradient matches a south–north gradient in rainfall in the KNP (Venter 1990), suggesting that rainfall regime may be an important driver of the spatial distribution of species diversity in the KNP (see also Linder 1991). In addition, other factors such as topographic heterogeneity might also contribute significantly to structuring spatial pattern of species richness in South Africa (Thuiller et al. 2006), and possibly in the KNP, although we did not evaluate them here.
Exclusion of Large Herbivores Results in Impoverished Species Communities
There is a concern that large herbivores (especially elephants) impact negatively on plant biodiversity in the KNP (Levick & Rogers 2008), but that there may be a critical density of large herbivores below which negative impacts are not felt (Baxter & Getz 2005). Previous research attempted to evaluate the impact of browsing on vegetation structure by comparing the composition of woody species along transects passing through the same KNP exclosures analysed here (Levick & Rogers 2008; Asner et al. 2009); however, impacts on phylogenetic diversity (measured here as MPD) – a possible surrogate for functional diversity (Crozier 1997; see also Cadotte et al. 2009) – were not assessed. We showed that when megaherbivores are excluded, species diversity generally decreases (see also Kalwij et al. 2010), but changes in phylogenetic diversity varied by spatial location.
Letaba provides one notable example, where, despite the decline in number of species (Shannon and SR), herbivore exclusion leads to a significant increase in the mean phylogenetic distance (MPD) between co-occurring species. This increase in MPD reflects a restructuring of the vegetation towards a less species-rich community composed of more phylogenetically diverse members. We suggest the unusual trends seen within Letaba may, in part, be a consequence of the high species richness and strong phylogenetic clustering of the initial community structure. In addition, Letaba exclosure is distinct from the other exclosures in many ways. First, it is situated on the banks of a river (the Letaba River) which causes period floods altering local plant diversity (Parsons et al. 2005). Second, the singularity of the Letaba exclosure may also be linked to its localization within the Mopane Bioregion (Mucina & Rutherford 2006), which is floristically and physiognomically much more homogenous than the rest of the KNP.
Further differences between sites might be linked to the variation in the age of exclosure establishment or in the size class of large herbivores excluded. However, we found no evidence for a bias with age or exclosure type. More likely, variation in site level attributes, such as productivity, soil nutrients and herbivore density can account for site differences in community composition after exclusion of large herbivores. For example, Olff & Ritchie (1998) and Bakker et al. (2006) suggested that in high-productivity environments, herbivore exclusion will generally decrease plant diversity whereas the opposite will be true in low-productivity environments.
We are limited in our study to data from five long-term exclosures, originally established several decades ago. We therefore interpret our results conservatively, and it is possible that differences in exclosure age and/or large herbivore exclusion might emerge with greater sample size. We strongly recommend that efforts are made to establish new exclosures within the KNP for future research. However, given current rates of environmental change and species loss, we cannot afford to wait several additional decades before making any attempts to analyse environmental trends. These established long-term experimental exclusion plots are the best data currently available, and our main results are clear: excluding large herbivores decreases plant diversity (Shannon and SR) but shifts in phylogenetic community structure are mixed.
Impacts of Large Herbivores on Community Structure may be Contingent on Initial Conditions
If large herbivores drive community clustering, we would then expect communities in exclosures to be largely overdispersed (NRIexclosure < NRIKNP). In addition, because some large herbivores can access the partial exclosures, we could also expect that partial exclosures would demonstrate intermediate clustering (intermediate between full exclosures and KNP) weaker than the KNP plots with unrestricted herbivore access (NRIPartial exclosure < NRIKNP), but greater than the plots in full exclosures (NRIPartial exclosure > NRIFull exclosure). However, we did not find evidence for a significant difference in community structure between partial and full exclosures; rather, our results indicate that impacts of large herbivore exclusion on community structure are contingent upon the initial community structure. Where communities were initially largely overdispersed (i.e. in the north of the Park), excluding large herbivores shifted communities to become more clustered. By contrast, where communities were initially more clustered (i.e. in the centre of the Park), the exclusion of large herbivores shifted community structure towards overdispersion.
Our results of conditional responses of communities to herbivore exclusion coupled with absence of clear differences between partial and full exclosures depart from the theoretical predictions that provided the stimulus for this study. These findings make it difficult to predict clearly how specialist vs. generalist herbivores impact on community structure in real environments. The differences observed between predictions and observed patterns probably reflect the complexity of ecological interactions in natural ecosystems. The restricted number of exclosures prevents us from drawing strong conclusions on the underlying causes of community rearrangement. For example, NRI was depressed in Letaba, but this happens to be the only site in the middle of the park and is a partial exclosure that has been up for only 8 years. By contrast, NRI was elevated in Nwashitsumbe, which is the only site in the north, and is a full exclosure that has been up for 43 years. It therefore remains possible that the observed shifts in community structure reflect some interactions between environment, exclosure age and herbivore exclusion.
Implications for Conservation
Our results are critical for predicting impacts on southern African ecosystems as large herbivores decline across much of the continent (Craigie et al. 2010) but increase in abundance locally, for example, within protected reserves. Critically, under herbivory might be as damaging to KNP plant communities as over herbivory; specifically, as large herbivores are lost from this ecosystem, we predict a subsequent reduction in plant species diversity. In addition, our study shows not only that large herbivores are key to maintaining woody plant diversity, but also that they may impose a specific phylogenetic structure on plant communities. Shifts in phylogenetic structure might have important consequences for community functioning because phylogenetic diversity can capture genetic and functional diversity, representing options in an uncertain future (Forest et al. 2007) and has been linked with ecosystem productivity (Cadotte et al. 2009; Flynn et al. 2011). However, the exclusion of large herbivores has mixed effects on community structure, dependent upon the initial community composition. Results were similar for both partial and full exclosures, suggesting that it is the largest herbivores (i.e. elephants and giraffes) that are responsible for driving these changes.
Characterizing the phylogenetic structure of communities will help in predicting community responses to ongoing environmental change. As we observed that community responses to herbivory varied across sites, we might also expect different community responses to future environmental change, including across locations in the KNP. As such, increasing the use of DNA-based community phylogeny analyses in African protected areas is likely a promising avenue for best management strategies in a changing world.
We thank the Royal Society (UK), South African National Research Foundation, European Commission, British Natural Environment Research Council, International Development Research Centre (Canada), Leverhulme Trust, the Government of Canada through Genome Canada and the Ontario Genomics Institute (2008-OGI-ICI-03) and the University of Johannesburg for financial support; SANPARK and the KNP Scientific Services for issuing research permits and helping with the logistics; Chabi Djagoun and Sadie Ryan for helping with the map and Thomas Rikombe, Roby Briden, Izack Smit, Thembi Khoza, Patricia Khoza, Cynthia Motsi, Herman van der Bank, Martyn Powell, Florent Borderie, Richard Greenfield, Guin Zambatis, Ranson Thethe, Roy Bengis, Holger Eckhardt, Barnabas Daru, Philipp Rousseau, Sylvie Duthoit, Renaud Lahaye, Velly Ndlovu and late Tabi Mhlongo for field assistance. Four anonymous referees and the editor provided valuable comments on earlier version of this manuscript.
Conflict of interest
The authors have no conflict of interest to declare.