Oecologia

, Volume 154, Issue 3, pp 561–570

Overdispersion of body size in Australian desert lizard communities at local scales only: no evidence for the Narcissus effect

Authors

    • Department of Ecology and Evolutionary BiologyCornell University
    • Cornell Lab of Ornithology
  • Julian Reid
    • Fenner School of Environment and SocietyAustralian National University
    • CSIRO Sustainable Ecosystems
  • Mark A. Cowan
    • Department of Environment and Conservation
  • Jeff Foulkes
    • Department for Environment and Heritage
Community Ecology

DOI: 10.1007/s00442-007-0849-1

Cite this article as:
Rabosky, D.L., Reid, J., Cowan, M.A. et al. Oecologia (2007) 154: 561. doi:10.1007/s00442-007-0849-1

Abstract

Both local and regional processes may contribute to community diversity and structure at local scales. Although many studies have investigated patterns of local or regional community structure, few have addressed the extent to which local community structure influences patterns within regional species pools. Here we investigate the role of body size in community assembly at local and regional scales in Ctenotus lizards from arid Australia. Ctenotus has long been noted for its exceptional species diversity in the Australian arid-zone, and previous studies have attempted to elucidate the processes underlying species coexistence within communities of these lizards. However, no consensus has emerged on the role of interspecific competition in the assembly and maintenance of Ctenotus communities. We studied Ctenotus communities at several hundred sites in the arid interior of Australia to test the hypothesis that body sizes within local and regional Ctenotus assemblages should be overdispersed relative to null models of community assembly, and we explored the relationship between body size dispersion at local and regional scales. Results indicate a striking pattern of community-wide overdispersion of body size at local scales, as measured by the variance in size ratios among co-occurring species. However, we find no evidence for body size overdispersion within regional species pools, suggesting a lack of correspondence between processes influencing the distribution of species phenotypes at local and regional scales. We suggest that size ratio constancy in Ctenotus communities may have resulted from contemporary ecological interactions among species or ecological character displacement, and we discuss alternative explanations for the observed patterns.

Keywords

Character displacementCommunity assemblyCompetitionNull modelRegional diversity

Introduction

Despite several decades of study, the processes influencing the assembly of ecological communities at local and regional scales remain a subject of considerable debate (Brown et al. 2000; Stone et al. 2000; Morena 2006). At local scales, niche differences among species have long been assumed to play a central role in community assembly because competitive exclusion and ecological character displacement can result from interspecific competition for shared resources (MacArthur and Levins 1967; Schluter 2000). The view that niche differences can promote species coexistence by reducing the intensity of interspecific competition is supported by theoretical and experimental results (MacArthur and Levins 1967; Schoener 1983a) and by patterns of trait overdispersion within communities, whereby traits related to resource use show greater dispersion among co-occurring species than expected in the absence of interspecific competition (Case and Sidell 1983; Schoener 1983b; Gotelli and Graves 1996).

Despite a large and growing body of literature on community-wide trait overdispersion (Dayan and Simberloff 2005), few studies have simultaneously considered the processes influencing community assembly at both local and regional scales (Stone et al. 2000). This is an important issue because the interpretation of patterns within local communities may be conditional upon patterns within the larger regional species pools from which local communities are drawn. For example, many studies have contrasted the dispersion of morphological or ecological traits within local communities to those of communities drawn randomly from the regional species pool (e.g. Fox and Brown 1993; Nipperess and Beattie 2004). Colwell and Winkler (1984) noted that such studies might underestimate the intensity of interspecific competition if the distribution of traits within regional species pools has already been influenced by competition. This phenomenon, termed the Narcissus effect, can occur if interspecific competition has caused the regional elimination of species most susceptible to competition.

To date, evidence for the Narcissus effect has been controversial. Brown et al. (2000) contend that regional assemblages of North American desert rodents have been shaped by competition. Stone et al. (2000) claim that little evidence exists for such structure at the level of the regional species pool that cannot be explained by the geography of speciation, differential dispersal abilities or other processes. They further suggest that only Bowers and Brown’s (1982) work on body mass in regional and local assemblages of heteromyid rodents supports the view that local processes can result in structured regional assemblages. Other studies employing detailed analyses of trait or functional group dispersion at local and regional scales have failed to reach consensus on this issue. For example, Nipperess and Beattie (2004) found evidence for morphological overdispersion of regional pools in arid Australian ant communities, but Gotelli and Ellison (2002) found that body sizes in North American ant communities were random or weakly clustered at regional scales.

Here we test for overdispersion of body size at local and regional scales in communities of Ctenotus lizards from arid Australia. With nearly 100 described species, Ctenotus is Australia’s most diverse terrestrial vertebrate genus. Ctenotus occurs in virtually all habitats in Australia, but the majority of species inhabit the arid and semi-arid regions of the continent, with diversity peaking in the western arid interior (Pianka 1981; James and Shine 2000). Body size in Ctenotus appears to be correlated with variation along several major niche axes, including prey size and microhabitat (Pianka 1969b, 1986; James 1991; Garland and Losos 1994), suggesting that size may be a useful summary of niche use in several dimensions. Although Ctenotus has served as a model system for understanding species richness and coexistence, there is no consensus on the significance of divergence along these or other resource axes for the assembly and maintenance of Ctenotus communities (Pianka 1989, 1969b; Morton and James 1988; James 1991; James and Shine 2000).

Because Ctenotus body size is putatively associated with niche use, overdispersion of body size at the local or regional scale may reflect interspecific competition or ecological character displacement. We use data drawn from several hundred Ctenotus communities within seven regional species pools to address two principal objectives. We test the hypothesis that body sizes within local Ctenotus assemblages should be overdispersed relative to those expected under null models specifying an absence of deterministic effects on body size. In addition, we evaluate evidence for a Narcissus effect by testing whether regional species pools are themselves overdispersed, and we investigate the relationship between the magnitude of overdispersion at local and regional scales.

Materials and methods

Community data

We surveyed Ctenotus communities at 283 sites across arid Australia between 1987 and 2005. Sites were nested within nine regional surveys, allowing for the delimitation of both regional assemblages and local communities (Fig. 1; Electronic Supplementary Material S1). Sites were generally selected to encompass the diversity of habitat types present within each region and were separated by several hundred to several thousand meters. At each site, a grid or linear array of 9–60 pitfall traps was established by burying 20-L buckets or PVC pipes in the soil, with each site sampling a total area of less than 1 ha. Traps within sites were connected using a continuous barrier of drift fencing, which directed lizards into the traps. Both the number of traps and their spatial arrangement were identical for sites within regions, but varied somewhat among regional surveys. All sites belonging to a particular regional survey were sampled for the same number of days, but the timing and duration of sampling varied among regional surveys (Electronic Supplementary Material S1). Pitfall traps were checked at least once daily during each sampling period, and snout-vent lengths (±1 mm) were measured for captured lizards. Although the majority of animals were released, representatives of each species from each regional assemblage were deposited in collections at the South Australian Museum and Western Australian Museum, where we used them to verify field identifications. Summaries of capture statistics for each dataset are given in Electronic Supplementary Material S1. A detailed overview of pitfall sampling methodologies employed in the collection of a representative dataset can be found in Reid et al. (1993).
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-007-0849-1/MediaObjects/442_2007_849_Fig1_HTML.gif
Fig. 1

a Location of nine datasets analyzed in this study. A Goongarrie Station (dataset abbreviation GG), B Lake Mason Station (LM), C two datasets from Lorna Glen Station (L1 and L2), D two datasets from Uluru National Park (U1 and U2), E MacDonnell Ranges (MR), F Lander River (LA), G Sangster’s Bore (SA). b Distribution of Ctenotus species richness across all sites (n = 283) within the nine datasets. A total of 21 species were detected overall, and six species were common to at least seven of the nine datasets. Capture and sampling statistics for each dataset are given in Electronic Supplementary Material S1

Body size estimation

We tested for overdispersion in the adult modal snout-vent length (SVL) of co-occurring species. Simply taking the average SVL of all captured lizards would have biased our estimates of adult body size due to the large but variable proportions of juvenile and sub-adult lizards captured during surveys. For each species, we plotted histograms of the SVLs of all captures; histograms typically showed a pronounced trough between juvenile and adult lizards (Fig. 2). This effect occurs because our sampling took place in the spring and autumn, when populations are a multimodal mixture of juvenile, subadult and adult lizards.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-007-0849-1/MediaObjects/442_2007_849_Fig2_HTML.gif
Fig. 2

Frequency histogram of snout-vent lengths (SVL) for all measured Ctenotus leonhardii individuals from Lorna Glen Station (datasets L1 and L2). Solid line indicates corresponding density estimated using kernel density estimation. The upper mode of the estimated density function was selected as an approximation of the “average” adult body size for this species (SVL = 69.13 mm)

To approximate modal adult body size, we used kernel density estimation to fit a continuous distribution to the set of SVLs for each species within each dataset (Silverman 1986). This method approximates the probability density function of a random variable and has been used in a variety of ecological applications (e.g. Manly 1996). We conducted kernel density estimation using the function "density" in the R statistical package (R Development Core Team 2004; http://www.cran.r-project.org/). The method is non-parametric because the data are not assumed to follow a known probability density function; rather, the shape of the estimated density function is determined entirely by the data. The density of body size at any point k is calculated as
$$ p(k) = \frac{1} {N}{\sum\limits_{i = 1}^N {W(k - X_{i} )} }, $$
where N is the number of observations, X are the data and W is a smoothing kernel, typically a probability density function. We used a Gaussian smoothing kernel for our estimates; thus,
$$ W(k - X_{i} ) = \frac{1} {{h{\sqrt {2\pi } }}}\exp {\left( {\frac{{(k - X_{i} )^{2} }} {{2h^{2} }}} \right)}. $$

This is equivalent to representing each data point Xi by a normal distribution with μi = Xi and with a common variance h2 for all data points. The density estimate p(k) is simply the summed densities of all points in X taken at point k. The standard deviation h of the smoothing kernel, also known as the bandwidth, controls the shape of the estimated density function. If h2 is too large, the estimated density p(k) will approach a uniform distribution; if it is too small, a mode will occur at every data point. We used the default bandwidth selection algorithm in “density”, which follows guidelines in Silverman (1986); density estimates were visually compared to histograms of the raw data to ensure that this procedure resulted in the appropriate choice of bandwidth.

The upper mode of the estimated density function was used as our index of adult body size (Fig. 2). Because body size in Ctenotus can vary geographically and because our datasets span a vast portion of interior Australia, we estimated region-specific sizes for species. Due to the close geographic proximity of datasets U1 and U2 (Uluru National Park) and of datasets L1 and L2 (Lorna Glen Station), we estimated body sizes from pooled capture data within each of these regions. We did not distinguish among male and female lizards because Ctenotus species show low levels of sexual size dimorphism (Pianka 1969b) and because they are often difficult to sex accurately in the field (Read 1998). Estimates of SVL for all species, as well as numbers of measured lizards per dataset, can be found in Electronic Supplementary Material S2.

Analysis of size overdispersion

We tested whether local Ctenotus communities showed overdispersion of body size using two metrics of size overlap. For each site, we ranked the set of co-occurring species by body size (SVL) and calculated the difference in log-transformed size between adjacent species. Applying a log-transform to body sizes ensures that these (log) size differences reflect size ratios among species, rather than absolute differences in body size (Gotelli and Graves 1996). Thus, a three-species community would be characterized by the ratios of the largest to intermediate-sized species and of the intermediate to smallest species. We determined the variance in size ratios (VAR) among co-occurring species at each site (Holmes and Pitelka 1968). This index is a property of all species present in the community and, hence, tests for community-wide overdispersion of body size; previous work has shown this metric to have desirable statistical properties (Pleasants 1994). Because at least two size ratios are required for VAR, we could only apply this metric to sites with three or more species of Ctenotus present. Observed indices for each site were compared to distributions of VAR for randomly assembled communities generated under three different null models:
  • Log-uniform model: Under the log-uniform model (LU), random assemblages were generated by sampling from a log-uniform distribution on the interval defined by the largest and smallest species occurring in a particular regional assemblage. A single iteration of the process for the ith site in the kth regional assemblage proceeded as follows. If N species occurred in region k, the endpoints of the distribution were fixed at the largest (max) and smallest (min) species observed in k. Two species were assigned observed min and max values, and body sizes for the remaining N − 2 species were sampled from a log–uniform distribution bounded by these min and max values. If ni species occurred at site i, ni body sizes were chosen from this source pool of N random body sizes in assemblage k.

  • Log-normal model: Under the log–normal (LN) model, body sizes for each site were drawn from a log–normal distribution with the same mean and variance as the observed distribution of body size for each region. As for the LU model, we chose exactly ni random body sizes for each site with ni species present.

  • Pool model: In this model, we compared observed assemblages to all possible combinations of the same species richness from the regional source pool (Schoener 1983b). Thus, body sizes at each site were constrained to those found in a particular region.

These three null models differ in the assumptions they make about the evolutionary and ecological processes underlying the distribution of size ratios in local communities. The LU null model assumes that all body sizes are possible within the morphospace defined by the largest and smallest species in the regional dataset, and the LN model assumes that body sizes follow a log–normal distribution. In contrast, the pool model assumes that the only possible body sizes at a given site are those already present in the regional species pool. If regional species pools are morphologically overdispersed, local communities drawn randomly with respect to body size from the regional pool will generally be overdispersed under LN and LU models, but not under the pool model.

We compared the observed VAR indices for each site to a null distribution tabulated from 2000 random assemblages generated under each null model. All simulations were conducted in the R programming environment (source code available from D. Rabosky). A meta-analysis framework was used to determine the significance of size overdispersion across all regional assemblages (Gurevitch et al. 2001; Gotelli and Ellison 2002). For each site, we calculated the standardized effect size (SES) for VAR as
$$ {\text{SES}} = \frac{{{\left( {I_{{{\text{obs}}}} - I_{{{\text{sim}}}} } \right)}}} {{s_{{{\text{sim}}}} }}, $$
where Iobs is the observed value of VAR, Isim is the mean of the simulated values for the site and ssim is the standard deviation of the simulated values. We used one-tailed, one-sample t tests to test our prediction that the distribution of SES values for VAR should be less than zero, which reflects relatively constant size ratios among co-occurring Ctenotus species.
For each null model, t tests were performed on the set of sites within each regional assemblage, and we combined P values across the nine datasets using the weighted Z method (Whitlock 2005). This approach has greater power and precision than other methods of combining P values (Whitlock 2005). In the weighted Z method, each independent test is assigned a weight, wi, and a weighted standard normal deviate is computed as:
$$ Z_{w} = \frac{{{\sum\nolimits_{i = 1}^k {w_{i} Z_{i} } }}} {{{\sqrt {{\sum\nolimits_{i = 1}^k {w^{2}_{i} } }} }}} $$
where Zi is the standard normal deviate from the ith independent test, and the weight wi is the corresponding df.

Regional scale

We repeated our analyses under the LU and LN models for each of the seven regional assemblages. We did not apply the pool model at the regional scale as many species were common to more than one region and because we used region-specific estimates of body size. As for the local scale, we used one-tailed t tests to test the hypothesis that SES values for regions should be less than zero.

If Ctenotus communities are characterized by a Narcissus effect, it should be most difficult to detect overdispersion of body size at the local scale when regional species pools already show evidence of size ratio constancy. Thus, a local assemblage might show no evidence for body size overdispersion relative to a community drawn randomly from the regional pool, if the regional pool is already morphologically overdispersed. We predicted that a Narcissus effect would result in an inverse correlation between the intensity of overdispersion at regional and local scales: as the intensity of regional-scale overdispersion increases, we expect a corresponding decrease in the intensity of local-scale overdispersion, at least under the pool null model. We tested this prediction by examining the correlation between regional effect sizes (SES) and the mean SES for sites within regions.

Habitat-specific patterns

Because arid zone Ctenotus communities vary considerably among habitats (Pianka 1986), we sought to determine the extent to which patterns observed across all sites differed among major habitat associations. We classified sites based on the abundance of spinifex grass (Triodia spp.), which is perhaps the single-most important habitat variable influencing Ctenotus community composition and diversity at local scales (Pianka 1986; Morton and James 1988; James and Shine 2000). Spinifex is a dense, prickly form of vegetation that can be exceedingly abundant in the western and central regions of the arid interior. It may provide lizards with protection from predators, favorable microclimates and an abundance of termites and other arthropod prey (Pianka 1981, 1986; Morton and James 1988); for these reasons, spinifex has been hypothesized to play a central role in promoting high species diversity in Australian desert lizard communities (Morton and James 1988; Pianka 1989). Our habitat categorization was based on a simple qualitative assessment of whether spinifex constituted the dominant form of vegetation at each site; we then tested for significant differences in SES values among these two habitat categories using two-sample t tests.

Results

Local scale

At the local scale, Ctenotus assemblages from arid Australia are characterized by community-wide overdispersion of body size, as assessed by the variance in size ratios among co-occurring species. One-tailed t tests revealed significant overdispersion of VAR in six datasets under the LU model, five under the LN model and three under the pool model. When Type I error probabilities for each dataset are combined, observed VAR indices are significantly lower than expected under LN, LU and pool models (Table 1). Nonetheless, effect sizes overall are not large, and we observed the smallest effect sizes under the pool model (Table 1).
Table 1

Meta-analysis of the dispersion of body size across sites in nine datasets as measured by the variance in size ratios within local assemblages of Ctenotus lizards from arid Australia

Modela

Sites

SESb

OVERc

UNDERd

Zw

P

Bonferroni Pe

LN

101

−0.193

6(5)

3 (1)

−5.183

<0.001

<0.001

LU

101

−0.289

7(6)

2 (1)

−6.347

<0.001

<0.001

Pool

101

−0.136

7(3)

2 (1)

−2.46

0.007

0.021

aLN, Log-normal; LU, log-uniform

bSignificance of effect sizes (SES) for each dataset and null model were assessed by one-sample t tests of the null hypothesis that body sizes are not overdispersed; the resulting P values for each dataset were combined using the weighted Z method

cOVER gives the number of datasets for which the mean of the metric indicated overdispersion of body size relative to the null model, with the number of datasets significant at P < 0.05 given in parentheses

eUNDER gives the number of datasets showing a pattern of underdispersion; the number in parentheses is the number of datasets for which the calculated t statistic exceeded the 95th percentile of the null distribution. Dataset LA (Lander River) showed a significant clustering of body size after Bonferroni correction under all three null models

eBonferroni probabilities are corrected for the three tests in this table

Because our nine datasets came from seven regions (U1 and U2 from Uluru National Park, and L1 and L2 from Lorna Glen Station), it could be argued that these datasets cannot be treated independently, and that we should have performed t tests only on sites within each region. To address this concern, we created pooled Uluru and Lorna Glen species-by-site matrices and repeated our analyses on the seven regional datasets. Results using these pooled datasets were virtually identical to those presented in Table 1 (combined, Bonferroni-adjusted P values: LN model P < 0.001; LU model P < 0.001; pool model P = 0.040).

A Kruskal–Wallis rank-sum test revealed significant heterogeneity in effect sizes among datasets (χ2 > 23.4; df = 8; Bonferroni P < 0.01 for all three null models). Although most datasets tended towards size ratio constancy at the local scale, dataset LA from the Lander River region of the Tanami Desert (Fig. 1, F) showed significant clustering of body sizes under all three null models (Fig. 3; Table 1 UNDER). If Lander River sites are excluded from the analysis, effect sizes do not differ among the remaining eight datasets (χ2 < 10.6; df = 7; Bonferroni > 0.48 for all three null models). Observed VAR indices and species richness for each site, as well as effect sizes under all null models, can be found in Electronic Supplemental Material S3.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-007-0849-1/MediaObjects/442_2007_849_Fig3_HTML.gif
Fig. 3

Standardized effect size (SES) of the variance in size ratios (VAR) for nine datasets (see Fig. 1 for abbreviations) considered in this study. Results shown for each dataset are the average SES values for the log-normal (LN), log-uniform (LU) and pool null models. Error bars indicate the standard error of these average SES values and thus reflect consistency of SES estimates for each dataset across the three null models, rather than the consistency of SES values among sites within datasets. Open squares indicate datasets for which SES values under at least two of three null models were significantly greater than or less than zero (P < 0.10; two-tailed). Most datasets tend towards size ratio constancy under at least two null models, but dataset LA is characterized by a clustering of size ratios

Regional scale

In contrast to patterns observed at the local scale, Ctenotus body sizes within the seven regional species pools show no evidence of size ratio constancy, and effect sizes under both LU and LN null models are small (Table 2). We plotted mean effect sizes for all sites within each region under the pool model against the regional effect size under the LU and LN null models (Fig. 4). We found no significant relationship between body size dispersion at local and regional scales under either null model (|Spearman’s ρ| < 0.2 and P > 0.7 for both null models), indicating that the distribution of body size at regional scales does not predict patterns at local scales.
Table 2

Analysis of size-ratio dispersion at the level of the regional species pool

Model

SES

t

Regions

P

Bonferroni P

LN

−0.023

−0.103

7

0.461

0.922

LU

0.262

0.548

7

0.698

1

https://static-content.springer.com/image/art%3A10.1007%2Fs00442-007-0849-1/MediaObjects/442_2007_849_Fig4_HTML.gif
Fig. 4

Relationship between body size dispersion at the regional scale under the LU and LN null models and dispersion at the local scale (pool model). Effect sizes at local scales are averages across all sites in each region. There is no significant relationship between regional and local dispersion of body size under either null model (LU model: Spearman’s ρ = 0.18, P = 0.71; LN model: ρ = 0, P = 1)

Habitat-specific patterns

It is clear that the presence or absence of spinifex accounts for much of the variation in community composition among sites; the majority of captures for most species occurred in either spinifex or non-spinifex habitats, and many species are found almost exclusively in one of these habitat categories (Fig. 5a). We found no significant differences in effect sizes for VAR for sites classified by prevalence of spinifex (Fig. 5b). Effect sizes in both habitats are generally similar to those obtained for the analysis of all sites (Table 1), although there is a non-significant trend towards greater overdispersion in sites not dominated by spinifex.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-007-0849-1/MediaObjects/442_2007_849_Fig5_HTML.gif
Fig. 5

a Variation in capture frequencies for 21 species of Ctenotus for sites characterized by the presence or absence of spinifex grass (Triodia sp.), a major habitat variable. Bars represent the proportion of total captures for each species that occurred on spinifex (black bars) and non-spinifex (gray bars) sites. Sampling intensity of spinifex and non-spinifex sites was approximately equal overall (124 spinifex sites and 111 non-spinifex sites with at least one Ctenotus capture). Most Ctenotus species occur in either spinifex or non-spinifex habitats, and only several species regularly occur in both habitats. A single habitat category accounted for at least 75% of all captures for 19 of 21 species and for 90% of all captures for 13 of 21 species. b Average standardized effect sizes (SES) for sites categorized by presence or absence of spinifex grass (Triodia sp.). Shaded bars LU model, cross-banded bars LN model, open bars pool model. Error bars indicate 95% confidence intervals. There are no significant differences between SES values for sites with and without spinifex under any null model

Discussion

Body size dispersion at local and regional scales

Our results indicate that Ctenotus assemblages in arid Australia are characterized by community-wide overdispersion of body size at local scales (Table 1). Using data drawn from nine independent surveys and multiple sites within seven regional species pools, we found that the variance in size ratios among co-occurring species is less than expected relative to three different null models. To the extent that body size is functionally related to resource use in Ctenotus, our results are consistent with a role for interspecific competition in generating size-structured local assemblages over ecological or evolutionary timescales. We consider alternative explanations for this pronounced overdispersion effect in more detail below.

Two findings would support the operation of a Narcissus effect across the spatial and taxonomic scales considered in this study. First, we would expect to observe little evidence for trait overdispersion at the local scale relative to the “pool” null model, whereby communities are drawn randomly from the regional species pool. Second, we should find evidence that regional species pools are trending towards trait overdispersion. Because the Narcissus effect represents an underestimate of the role of competition inferred under “pool” null models caused by sampling a postcompetition regional species pool (Colwell and Winkler 1984; Stone et al. 2000), we further predicted that regional species pools trending towards morphological overdispersion would show reduced evidence for size ratio constancy at the local scale under the pool null model. This reduction in the magnitude of trait overdispersion was predicted because the distribution of phenotypes within regional pools would already have been shaped by species interactions.

Our results are inconsistent with a Narcissus effect. Although observed effect sizes at the local scale were greater under the continuous distribution null models (LU and LN), we nonetheless observed significant size ratio constancy under the pool model (Table 1). Furthermore, we found no evidence that regional species pools are trending towards overdispersion of body size (Table 2). Finally, there is no relationship between regional scale dispersion of body size and body size dispersion at the local scale under the pool model (Fig. 4).

Habitat-specific patterns

When we classified sites based on the abundance of spinifex, we found no significant differences in VAR between sites with and without spinifex cover (Fig. 5b), and effect sizes are comparable to those observed for all sites considered together (Table 1). We appreciate that this simple binary habitat classification fails to recognize many plant and substrate associations within the Australian arid-zone. For example, the spinifex category included sites characterized by many different canopy species occurring over a range of substrates, including sandplain with Eucalyptus gongylocarpa overstory, Thryptomene-dominated dune and swale and rubble-strewn erosional surfaces. Nonetheless, spinifex is clearly one of the most important habitat variables influencing Ctenotus community composition (Fig. 5a; Pianka 1969a, 1972), and our results suggest that size ratio constancy within Ctenotus communities is not driven by patterns associated with a single widespread habitat.

Evolutionary and ecological causes of size ratio constancy

The tendency towards size ratio constancy suggests that the distribution of Ctenotus phenotypes in local communities has been influenced by interspecific competition. Trait overdispersion can be caused by competitive exclusion occurring over ecological timescales or by character displacement (Case and Sidell 1983; Dayan and Simberloff 2005). Although we cannot rigorously demonstrate character displacement without incorporating additional phylogenetic information (Losos 1990), we note that body size is one of the principal phenotypic axes along which character displacement has been observed in lizards (Giannasi et al. 2000; Radtkey et al. 1997; Melville 2002).

Interspecific competition can influence trait distributions within communities by favoring phenotypes that exploit under-utilized resources, or through behavioral interference among individuals with similar phenotypes. The latter, typically referred to as interference competition, is an alternative to the more commonly considered interactions involved in direct interspecific resource competition (Case and Gilpin 1974; Adams 2004). Interference competition is particularly likely if the intensity of behavioral interference is proportional to the extent to which individuals overlap in body size (e.g. Palomares and Caro 1999). Consistent with the possibility of interference competition, we note that body size plays a critical role in mediating interspecific aggression in some lizards (Losos 1996; Melville 2002) and in a number of other taxa (Alatalo and Moreno 1987; Robertson 1996).

Such behavioral interactions do not rule out resource competition as a causal process, because size-conditional behavioral interference could reflect an evolutionary response to potential resource competitors. Resource competition has been implicated as a primary cause of interspecific aggression (Polis et al. 1989), and in some taxa, the intensity of interspecific aggression is related to dietary overlap (Donadio and Buskirk 2006). Melville (2002) attributed character displacement in several species of Tasmanian lizards to interference competition for basking sites, a limiting resource in the alpine habitats where these species occur. Considerable theoretical and empirical evidence now supports the view that interference competition can result in character displacement (reviewed in Dayan and Simberloff 2005), although recent work on carnivores suggests that some types of interference competition may actually favor convergence in body size (Donadio and Buskirk 2006). The relationship between interference and resource competition deserves further study in Ctenotus and other systems.

Alternative explanations for size ratio constancy that do not require a role for interspecific competition include non-random distributions of resources (Schluter and Grant 1984), predation (Jeffries and Lawton 1984) and reproductive character displacement (Pfennig and Pfennig 2005). Although overdispersion of Ctenotus phenotypes may be attributable to prey size or microhabitat features, characterizing differential resource availability in such complex, multidimensional systems poses a daunting challenge. There is no obvious multimodal signal in the distribution of prey sizes taken by Ctenotus (Pianka 1969b), possibly due to the large numbers of termites consumed by all species. However, it is possible that discontinuities exist along other resource axes or in prey size during periods of severe resource limitation. In one of the most thorough analyses of resource partitioning in Ctenotus, James (1991) found that five syntopic Ctenotus species partitioned termite prey by generic or functional-group identity, but not by size, during the driest year of his study.

It is difficult to assess the effects of selective predation on the distribution of body size in Ctenotus, but we note that Ctenotus predators include a variety of generalists, including several species of varanid lizards, elapid snakes and the pygopodid Lialis burtonis (Patchell and Shine 1986). Varanid lizards, for example, are a principal Ctenotus predator in Australian deserts, and dietary analyses indicate that they opportunistically prey upon virtually all species of Ctenotus that they encounter (Pianka 1994). As virtually nothing is known regarding mate choice in Ctenotus, we can at best acknowledge the possibility that reproductive character displacement in size might have contributed to the observed patterns.

If ecological character displacement underlies patterns of size ratio constancy within local Ctenotus communities, why would we have failed to observe overdispersion at the regional scale? One possible explanation is that trait overdispersion in Ctenotus at the local scale may be attributable to evolutionary shifts in habitat among similar-sized species. If the intensity of competition is related to body size overlap, natural selection might favor different patterns of habitat use, rather than evolutionary shifts in body size. Such similar-sized species would thus be retained in the regional species pool, and the regional pool would show little evidence for size ratio constancy. However, at the local scale, we would observe reduced co-occurrence among similar-sized congeners. Because our local communities were defined precisely to capture within-habitat patterns of co-occurrence, we consider this hypothesis to be worthy of future study.

Conclusion

Among terrestrial vertebrates, regional-scale trait overdispersion has been widely reported in communities of carnivorous and granivorous mammals (Dayan et al. 1989; Yom-Tov 1991; Dayan and Simberloff 1994; Kieser 1995; Ben-Moshe et al. 2001). Our large-scale surveys of arid Australian Ctenotus communities reveal a different view of community organization. Although Ctenotus communities are characterized by size ratio constancy at local, within-habitat spatial scales, we find little evidence that local processes in this system “feed up” (Brown et al. 2000) to influence the distribution of body size in regional species pools. Previous research has suggested that species richness at local scales in this system is primarily determined by processes occurring at the regional scale (James and Shine 2000), but our results do not support a similar relationship underlying the distribution of species phenotypes in local and regional assemblages.

Acknowledgments

We thank D. Adams, A. Agrawal, M. Austin, H. Greene, L. Joseph, I. Lovette, A. McCune, and S. Morton for comments on the manuscript and/or valuable discussion of these topics. For assistance with survey design and other logistics, we thank S. Argus, G. Armstrong, L. Baker, P. Doughty, S. Eldridge, E. Foster, N. Gambold, M. Gillam, J. Gillen, M. Hutchinson, the late K. Jones, I. Kealley, K. Kenneally, A. Kerle, S. Morton, N. de Preu, and D. Stefoni. In addition, we thank the many individuals who assisted with other aspects of this project, including G. Allen, J. Alley, JM Armstrong, M. Barritt, B. Barton, K. Bellchambers, A. Brook, R. Burton, S. Campbell, D. Carter, V. Clarke, J. Cole, J. Coulter, M. Coulter, A. Duguid, M. Fleming, B. Gardiner, K. George, D. Gibson, D. Graham, A. Grattidge, Greenie, G. Hearle, R. How, J. and J. Kavanagh, D. Langford, B. Lewis, R. Lynch, C. Martin, D. Mason, G. McKenzie, J. Patten, N. Rabillier, A. Richardson, D. Risby, B. Ryan, D. Schunke, P. Spencer, L. Stokes, P-J. Waddell, A. Willson, A. Woosnam, and Conservation Volunteers Australia. We also thank the traditional owners from Willowra, Yuendemu and Mutitjulu for allowing us to conduct field work at the Lander River, Sangsters Bore and Uluru Kata Tjuta National Park sites, respectively. Funding for this study was provided by NSF-OSIE- 0612855, an EAPSI Fellowship to DLR, administered jointly by the U.S. National Science Foundation and the Australian Academy of Science, and by Sigma Xi, the Evolutionary Biology Program at the Cornell University Lab of Ornithology, the Mario C. Einaudi Center for International Studies, the Goldfields and LANDSCOPE branches of the Western Australia Dept. for Environment and Conservation, and the Australian National Heritage Trust. Research conducted for this study conforms to current Australian law in every respect.

Supplementary material

442_2007_849_MOESM1_ESM.doc (34 kb)
ESM1 (DOC 34 kb)
442_2007_849_MOESM2_ESM.xls (19 kb)
ESM2 (XLS 19 kb)
442_2007_849_MOESM3_ESM.xls (26 kb)
ESM3 (XLS 27 kb)

Copyright information

© Springer-Verlag 2007