D.T. Bilton, School of Biological Sciences, University of Plymouth, Drake Circus, Plymouth, PL4 8AA, UK. Tel.: +44 1752 232902. Fax: +44 1752 232967. E-mail: email@example.com
1Nestedness is a composite property of many suites of biotas. Such patterns may be driven by dispersal limitation, species–area relationships, hierarchical niche requirements, or occur as an artefact of passive sampling. Despite its widespread occurrence, few studies have explored the factors underlying nested subset structure, and ecological distinctions between nested and non-nested (idiosyncratic) taxa within a region have been largely ignored.
2Macroinvertebrate assemblages from 45 heathland ponds in south-west England are used to (i) unravel the relative importance of processes driving nested subset structure and (ii) test spatially explicit hypotheses concerning the response of nested and idiosyncratic taxa to parameters shown to structure assemblage-level nestedness.
3Despite being dominated by taxa with good powers of inter-site dispersal, pond macroinvertebrate assemblages were found to be significantly nested. This nesting was not due to passive sampling, and was best explained by pond area, with habitat parameters and isolation being of secondary importance.
4The spatial responses of nested and idiosyncratic taxa matched predictions; nested taxa showed strong spatial structure, which was reduced when the effects of pond area and habitat were removed. In contrast, a greater proportion of idiosyncratic taxa were completely spatially random and exhibited weaker responses to factors that structure assemblage level nestedness.
5Nested and idiosyncratic species tend to differ ecologically: idiosyncratic taxa generally possess broad ecological tolerance and good dispersal capacity, whilst nested species are more likely to have narrow tolerances or limited powers of dispersal.
6Factors structuring nestedness in ponds can be viewed as probabilistic filters which act to limit the spatial distribution of species with narrow ecological tolerance or low dispersal tendency. Nestedness analysis alone fails to elucidate processes that structure assemblage composition. The additional use of spatially explicit analyses is important if processes that generate nested pattern across a region are to be understood.
Nested subset patterns could be caused by several factors. Passive sampling could generate nestedness as an artefact of underlying stochastic principles, as rare species are less likely to be sampled in a given area than common species (Gaston & Blackburn 2000; Fischer & Lindenmayer 2002). Habitat isolation also creates nested subsets through dispersal limitation, as species differ in their ability to colonize distant sites (Patterson & Atmar 1986). Additionally, area may drive nestedness because larger habitat patches support species with both large and small minimum area requirements, whilst smaller patches only support the latter (Boecklen 1997; Wright et al. 1998). Nested distribution of habitat types, disturbance regime and hierarchical niche relationships may also produce nested assemblages (Kolasa 1996; Honnay et al. 1999; Patterson & Atmar 2000). In contrast, frequent between-patch dispersal has been proposed to erode nested patterns, e.g. in aquatic invertebrates (Boecklen 1997; Wright et al. 1998), serving to homogenize assemblage composition.
In addition to indicating the presence of nested subset structure, nestedness analysis enables the recognition of significantly non-nested distributions due to species or habitat checker boarding (Gotelli & McCabe 2002), or spatial turnover (Gaston & Blackburn 2000). Where significant nesting does exist, species that conform to the overall assemblage nestedness pattern can be differentiated from taxa which depart from nestedness (known as idiosyncratic taxa), which occur more frequently than one would predict in species poor sites (Atmar & Patterson 1993, 1995).
Studies of assemblage structure, including nestedness analysis, often have inherent spatial components, and therefore demand the use of spatially explicit analyses (Wilson 1999). For example, Keitt et al. (2002) argue that the relative and absolute importance of environmental variables for species occurrence and abundance may be incorrectly assessed if spatial autocorrelation in their patterns is ignored. Despite the importance of accounting for spatial structure within analyses of assemblage composition, only a single study published to date has examined nestedness in a spatial context (Hausdorf & Hennig 2003). Factors that can structure nested subset patterns (e.g., habitat area and type) may be spatially autocorrelated in the landscape, and the occurrence of nested taxa might therefore show a similar pattern of autocorrelation. In contrast, idiosyncratic taxa, which depart from the nested pattern, might be expected to exhibit different spatial structure, showing either negative or random responses to factors that drive nestedness. If dispersal erodes nested structure, as suggested by Boecklen (1997), idiosyncratic taxa should tend to be species that are especially strong and active dispersers, and may therefore be more widely distributed and spatially random than nested taxa, which would tend to have locally clumped distributions. To date, these predictions have not been tested explicitly.
This study is the first to determine the relative importance of factors driving nested subset structure in a spatial context and compare the response of nested taxa to those which depart from this pattern. We use macroinvertebrate assemblages in heathland ponds to (i) unravel the processes that may underlie nested subset structure and (ii) examine the spatial responses of idiosyncratic and nested taxa to parameters shown to structure assemblage-level nestedness. Ponds are an ideal model system as they form habitat islands for aquatic species (Bilton, Freeland & Okamura 2001), and can vary extensively in their physical characteristics and the richness of their biota across small spatial scales (Kiflawi, Eitam & Blaunstein 2003). The fauna of small ponds is also dominated by mobile species, many of which are capable of dispersing between individual waterbodies repeatedly during their lives.
Materials and methods
Data on macroinvertebrate assemblage composition were generated for 45 heathland ponds on the Lizard Peninsula, south-west Cornwall UK. All ponds were fish-free, but varied substantially in area, permanence and vegetation composition, although all were above ultra-basic serpentine geology on heathland/unimproved grassland (see Bilton, Foggo & Rundle 2001 and Rundle et al. 2002 for details).
Invertebrates were sampled with a 1 mm mesh hand net (frame 20 × 25 cm) during February 2000, a time of year best suited to representative sampling of assemblages because most ponds in the region were wet, and macroinvertebrates abundant. Five standardized 1-m sweeps were stratified between beds of vegetation with different macrophyte species compositions, according to the relative frequency of stand types in each pond. Vegetation types were identified in the field on the basis of species composition and complexity, with 1–5 stand types being recognized in each pond. Most ponds were dominated by Glyceria and Juncus species, with Potamogeton, Ranunculus and Eleogiton in many of the larger sites. The sampling strategy employed in this study has been shown to consistently obtain the majority of invertebrate taxa present within ponds, allowing comparison of their assemblages (Foggo, Rundle & Bilton 2003). Sweeps were pooled and macroinvertebrates and detritus preserved in 70% alcohol. In the laboratory samples were sorted and animals identified to species, except for chironomids which were identified to genus.
pond chemistry, habitat, isolation and area
Two water samples were collected from each sampling station, in acid washed polypropylene bottles, for subsequent analysis of metal cation concentration. Mean water depth was recorded and pH readings were also taken on-site using a Solomat (Bishop's Stortford, UK) probe. Water hardness was calculated as 2·5[Ca2+]+ 4·1[Mg2+] (Gower et al. 1994). Macrophyte and semi-terrestrial vegetation species composition was also examined at each sampling station in late May of the same year; when most species were in flower and could be readily identified. Taxa present in the area from which invertebrates were sampled were recorded and identified to species, bryophytes and Callitriche spp. were noted but not identified further. The number of plant species recorded at each sampling station varied from 2 to 22 (mean 8·6, SE 0·76).
In order to examine the relationship between nestedness and habitat parameters a summary of vegetation and physicochemical variables was produced. Number of macrophyte species, mean depth, pH and water hardness were normalized and standardized, and subjected to Principle Components Analysis (PCA). First PCA axis score was then used as a simplified measure of pond habitat (Honnay et al. 1999) in subsequent analyses.
Accurate estimates of pond area were derived by using differential GPS (Trimble, Hook, UK) to map the margin of each pond; this process also generated central point co-ordinates for each pond, which were used to create a between-pond distance matrix. Such an approach was deemed appropriate because ponds were small relative to the spatial separation between them. Pond isolation was calculated as the sum of all pair-wise distances to other ponds (Jeffries 2003). Pond area ranged from 2 m2 to 15 005 m2 (mean 1047·9 m2, SE 409·8, median 38 m2). It was not possible to explicitly determine individual pond permanence in this study, although this is generally positively related to area (e.g. Williams 1987).
nested subset analysis
Nestedness was assessed using the nestedness temperature calculator (Atmar & Patterson 1993, 1995). The metric employed (T) has various advantages over other measures of nestedness including (i) matrix size independence, (ii) easy identification of idiosyncratic taxa (idiosyncratic species have above average temperatures), and (iii) simultaneous maximal nesting across species and sites (Patterson & Atmar 2000). The lack of stringency of the underlying null model used by the temperature calculator has been the subject of recent criticism, particularly because matrices generated by passive sampling have been shown to be significantly nested (Jonsson 2001; Fischer & Lindenmayer 2002). In order to test the significance of the observed nestedness more rigorously, and to discount passive sampling and species richness effects as sources of nestedness, we used two additional null models. To examine passive sampling effects, we created random matrices fixing the values for species’ overall occurrence to that in the observed matrix (Gotelli & Graves 1996). One hundred such matrices were generated and the nestedness temperature calculator was used to calculate the range of T-values expected from such random sampling. If passive sampling structures nestedness, the observed matrix temperature should lie within this expected distribution (Fischer & Lindenmayer 2002).
A second algorithm was then used to create an additional hundred matrices, fixing both species occurrence and number of species per site (Brualdi & Sanderson 1999). These null distributions were used to test the effects of species richness upon nestedness. If richness drives the nestedness pattern, the observed matrix temperature would again be predicted to lie within the expected distribution.
correlates of nestedness
To examine the effects of area, habitat and isolation (factors purported to drive nestedness in many systems) upon nestedness, we first calculated site nestedness order, using the matrix packing algorithm within the nestedness temperature calculator (Atmar & Patterson 1995). Second order partial correlation analysis (Sokal & Rohlf 1995) was then used to examine bivariate correlations between the nestedness order and area, isolation and habitat PCA scores. Partial correlations were performed on ranked data as area and isolation were skewed and transformation failed to normalize their distribution.
The relationship between nestedness and area, isolation and habitat factors was also investigated using an approach developed by Lomolino (1996). Sites were ranked by species richness, and the number of departures from nestedness quantified by recording the number of times the absence of a species was followed by its presence in the next most species rich site, giving a basic measure of internal nestedness (Honnay et al. 1999). The same procedure was then repeated on the matrix after it had been reordered by rank pond area, rank isolation, and rank habitat (lowest PCA score first), respectively. The observed number of departures for each of these rankings was then compared with the range of values gained from 1000 randomizations of the matrix. The matrix reorder variable resulting in the lowest number of departures is that which correlates best with observed nestedness structure.
nested vs. idiosyncratic taxa
To examine the spatial responses of taxa to correlates of nestedness, autocorrelation analyses were conducted using The R Package (Casgrain & Legendre 2001). Correlograms were constructed using 14 equal distance lag intervals, in accordance with Yule's rule (Casgrain & Legendre 2001), which determines the optimum number of lags based on the number of distance pairs in the data; inter-lag distance was 675 m with maximum inter-pond distance 9450 m. Correlograms of total, nested and idiosyncratic species richness were produced using Moran's I computed for each distance class, with significance of Moran's I at each lag corrected for multiple comparisons using the Bonferroni method. The effect of pond area and habitat were determined using additional correlograms of residuals for each of the three richness measures regressed against pond area and habitat PCA scores, respectively (P. Legendre, personal communication). If pond area and/or habitat strongly influence the spatial structure of richness, these correlograms should show significant changes over the originals and indicate a lack of autocorrelation; if pond area or habitat has little influence, excluding their effects should leave the correlogram relatively unchanged.
Relative abundance data for taxa occurring in three or more ponds were also examined for autocorrelation, using correlograms to compare spatial structure in nested and idiosyncratic species. The numbers of lag distances with significant positive or negative autocorrelation were summed across all nested species, and the mean values per taxon taken as a measure of typical spatial structure. The same procedure was then performed for idiosyncratic taxa. Finally, the effects of area and habitat on individual species’ abundance distributions were examined, again by plotting correlograms of regression residuals as described above.
Principle Component Analysis (PCA) showed that axis one accounted for 41·2% of variation in the pond habitat data (axis 1 eigenvectors: log depth −0·528; pH −0·459; log plant species richness −0·502; log water hardness −0·508). Low PCA axis one scores represent ponds which were relatively deep, with approximately neutral pH, high macrophyte richness and low water hardness, typical of larger more permanent sites.
nested subset analysis
The macroinvertebrate presence absence matrix had a temperature of 15·5° which was significantly nested when compared with all three null models (Table 1). Around a quarter (31/118) of taxa recorded were idiosyncratic in their distribution, with Coleoptera and chironomids making up 81% of these (Table 2), as opposed to 62% of nested species. Coleoptera were particularly well represented amongst the idiosyncratic taxa, comprising 58% of the total, as opposed to 39% of those which were nested. Partial correlation indicated that the proportion of idiosyncratic taxa per site was negatively correlated with pond area (rs = −0·695, P < 0·001), indicating that idiosyncratic taxa form a greater proportion of the total taxon richness in small ponds. The proportion of idiosyncratic taxa was also significantly correlated with the habitat PCA (rs = 0·494, P < 0·001), whereas the corresponding correlation with isolation was not significant. The absolute number of idiosyncratic taxa per pond was not significantly correlated with area, isolation or habitat.
Table 1. Observed and expected nestedness temperatures based on three different null models (i) default model, species occurrence and site species richness are equiprobable, (ii) species occurrence fixed to that observed, and (iii) species probability and site species richness fixed to that observed
(i) Default null model.Number of species occurrences and site species richness equiprobable (n = 1000)
Mean 56·06°; SD 1·94° Range 50·0–63·0°
(ii) Passive sampling effect.
Mean 34·25°; SD 1·32°
Number of species occurrences fixed and site species richness equiprobable (n = 100)
(iii) Species richness effect.
Mean 17·99°; SD 0·37°
Number of species occurrences fixed and site species richness fixed (n = 100)
Table 2. Idiosyncratic taxa that are less nested than average, having temperatures greater than 15·5°
Occurrence (no. ponds)
Occurrence (no. ponds)
Agabus bipustulatus (L.)
Ilybius montanus (Stephens)
Anacaena lutescens (Stephens)
Dryops striatellus (Fairmaire & Bristout)
Metriocnemus van der Wulp
Gyrinus substriatus Stephens
Graptodytes flavipes (Olivier)
Haliplus lineatocollis (Marsham)
Paratanytarsus Thienemann & Bausse
Haliplus fulvus (Fabricius)
Helophorus aequalis Thomson
Limnephilus vittatus (Fabricius)
Helophorus brevipalpis Bedel
Helophorus grandis Illiger
Corixa punctata (Illiger)
Helophorus minutus Fabricius
Corixa affinis Leach
Helophorus obscurus Mulsant
Hydroporus melanarius Sturm
Lymnaea truncatula (Muller)
Hydroporus planus (Fabricius)
Hydroporus pubescens (Gyllenhal)
Crangonyx pseudogracilis Bousfield
Hydroporus tessellatus Drapiez
Ochthebius dilatatus Stephens
Enallagma cyanthigerum (Charpentier)
correlates of nestedness
Partial correlation indicated that nestedness order correlated with pond area (rs = −0·460; P < 0·01) as it also did when employing Lomolino's (1996) method (Table 3). Nestedness order was also significantly related to pond isolation using Lomolino's technique (Table 3), but not using partial correlation. Partial correlation additionally indicated a significant relationship between nestedness and habitat PCA (rs = 0·336; P < 0·05).
Table 3. Lomolino (1996) departures for matrices reordered according to pond area, isolation and habitat PCA score compared with 1000 randomizations of site order
Number of departures, D
Number of randomizations giving D < observed
Sites ranked by species richness
Sites ranked by area
Sites ranked by mean isolation
Sites ranked by habitat PCA score
Sites ranked randomly (n = 1000)
Mean 581·9, SD 22·64
nested vs. idiosyncratic species
The correlogram of total species richness (Fig. 1a) indicates that the total number of taxa was significantly structured through space (nine significant lag distances). The correlograms of the area and habitat PCA regression residuals were more spatially random, with only three significant lag distances each. This indicates that area and habitat are significantly structuring the spatial response of total species richness; this is particularly clear at low lag distances (675 m to 4050 m; Fig. 1a).
Richness of nested species shows a similar but stronger pattern to that for total species richness (Fig. 1b); with 10 significant lag distances, whilst the correlograms of area and habitat residuals have only three. Idiosyncratic species richness (Fig. 1c) shows weaker spatial structure, with four significant lags; comparison of this correlogram with those of the habitat and area residuals reveals little change.
On average individual idiosyncratic species showed less evidence of significant spatial structuring than nested species (Table 4) with means of 1·08 (n = 24) significant lag distances per taxon compared with 2·04 (n = 54; one tailed Mann–Whitney test, W = 736, P < 0·01). The mean number of negative lags was significantly greater for nested than for idiosyncratic taxa (1·43 compared to 0·67; W = 692, P < 0·01). No significant difference in the number of significant positive lag distances was observed between idiosyncratic and nested taxa. The number of macroinvertebrate species that were completely spatially random (i.e. random at all lag distances) represented a greater proportion of idiosyncratic taxa (54·2%) than nested taxa (24·1%).
Table 4. Comparison of idiosyncratic and nested species spatial structure. Proportion of taxa showing completely spatially random (CSR) distributions and mean number of spatial lags per taxon showing significant autocorrelation after Bonferroni correction (P < α/14, where α = 0·05). Asterix indicates that idiosyncratic taxa have fewer significant lag distances than nested taxa (one tailed Mann–Whitney P < 0·01)
No. taxa occurring in ≥ 3 ponds
Proportion of taxa that are completely spatially random (CSR)
Mean total no. of lags per taxon showing significant autocorrelation
Mean no. of lags per taxon showing significant negative autocorrelation
Mean no. of lags per taxon showing significant positive autocorrelation
The spatial responses of individual species to pond area and habitat form a continuum (Fig. 2). Overall, idiosyncratic taxa (e.g., Dryops striatellus (Fairmaire & Bristout), Fig. 2c) were more random in their spatial distribution and showed less response to pond area and habitat characteristics than nested taxa (e.g., Dryops luridus (Erichson), Fig. 2a) which showed stronger spatial autocorrelation. However, many idiosyncratic and nested species showed an intermediate level of response (e.g., nested Dryops auriculatus (Geoffroy), Fig. 2b).
This study shows that local assemblages within a region can show significant levels of nestedness despite being dominated by taxa with good powers of inter-locality dispersal (Rundle et al. 2002). This nested structure was not due to passive sampling or directly related to species richness (Tables 1 and 3).
Boecklen (1997) and Wright et al. (1998) showed that aquatic invertebrate assemblages exhibit lower degrees of nested subset structure than other taxonomic groups. They infer that high rates of dispersal amongst habitat islands might mask nested subset pattern by increasing the spatial turnover of species. This study suggests that a high level of inter-site dispersal does not always preclude the presence of nestedness in aquatic invertebrate systems. Significant nested subset structure has been shown for other taxonomic groups with high interpatch dispersal, for example butterfly assemblages at both large and small spatial scales (Fleishman & MacNally 2002; Fleishman et al. 2002; Summerville, Veech & Crist 2002).
Both partial correlation and Lomolino's technique show area to be the best correlate of nestedness, although pond habitat and isolation were also important. All three of these interrelated factors are likely to act together to shape nestedness. Large ponds with low habitat PCA scores (i.e. approximately neutral pH, higher macrophyte species richness with greater depth) and that are close to other ponds unsurprisingly tend to be the most species rich, and are basal to a pattern of nested pond assemblages throughout the landscape. Small sites with higher habitat PCA scores have lower total species richness, but support assemblages that contain a similar number of idiosyncratic taxa to that found in large ponds.
Patch-area dependent extinction processes are reported to shape nestedness when area correlates well with the observed pattern (Atmar & Patterson 1993; Wright et al. 1998; Honnay et al. 1999). This is particularly applicable for fragmented habitats where relaxation is occurring, and may similarly happen when ponds shrink as they dry. However, during February temporary pond habitat is at maximum extent, and small ponds may instead have been depauperate because they (i) provide less habitable space, (ii) have been wet for less time than larger water bodies, allowing less time for colonization, and (iii) are risk prone for taxa without suitable adaptation to cope with or avoid drought.
Another factor that potentially structures nested subsets is the hierarchical distribution of niche space (Kolasa 1996). In this case, species well adapted to the temporary pond environment (usually referred to as temporary pond ‘specialists’) should in fact be ubiquitous generalists, and species more limited by pond hydro-period should be specialists occurring in a subset of sites. However, hierarchical niche relationships do not seem to be a major structuring force for nestedness in our system, as many species that could be considered generalists, e.g. Limnephilus vittatus (Fabricius), Helophorus brevipalpis Bedel and Corixa punctata (Illiger) (Table 2) are idiosyncratic. These generalists are distributed across ponds of different species richness, area, isolation and habitat type, but because they occur in species poor sites the nestedness temperature calculator model expects them to be present in all assemblages of greater species richness. They are therefore idiosyncratic because they have unexpected gaps in their distribution.
Patterns in the number of nested species within individual assemblages (Fig. 1a,b) are structured largely by the effects of pond area and habitat, whilst the number of idiosyncratic taxa is only weakly governed by pond characteristics (Fig. 1c). A similar effect is also evident in the spatial distribution of individual species. For instance, pond area and habitat are important in structuring the distribution of the nested water beetle Dryops luridus (Fig. 2a) but have little effect on the spatial distribution of its idiosyncratic relative Dryops striatellus (Fig. 2c). The pattern seen with individual taxa is sometimes less clear cut than that at the assemblage level, however, with a number of species such as the nested Dryops auriculatus (Fig. 2b) showing an intermediate response to pond area and habitat. Despite this continuum of response, nested taxa show greater spatial structure than idiosyncratic species as on average they have more significant negative spatial lags (Table 4). This indicates that nested taxa are more dispersed through the landscape, due to avoidance of unsuitable sites. In contrast the more random spatial distributions of idiosyncratic taxa indicate that they are not actively avoiding species rich sites but opportunistically colonize all types of pond.
The split into nested and idiosyncratic taxa in our study appears to relate to differences in life history strategy. Idiosyncratic species occur more frequently than expected in species poor sites, which in our study are small, ephemeral waterbodies. Compared with their nested relatives, idiosyncratic species tend to be active dispersers throughout adult life and/or possess adaptation to drought in one or more life stage, such as semi-terrestrial larvae (e.g., Helophorus spp.), amphibious behaviour (e.g., Lymnaea truncatula), short larval duration or aquatic larvae that can survive in moist mud (Williams 1987). These life-history characteristics allow such taxa to utilize small sites that fill during spring for reproduction (e.g., Helophorus brevipalpis, Hydroporus planus (Fabricius) and Agabus bipustulatus (L.); Fernando 1958; Landin & Stark 1973). Many of these species retain the ability to disperse throughout adult life and can track environmental change, dispersing to permanent refugia during the summer months (Pajunen & Jansson 1969; Landin & Stark 1973; Svensson 1998, 1999). In comparison nested taxa are less frequently found in highly temporary waterbodies, and show reduced ability and/or tendency to fly (Brown 1951).
Factors structuring nested subsets in ponds might be viewed as probabilistic filters (Wright et al. 1998) which act at the individual species level to limit the spatial distribution of species with narrow ecological tolerance or low dispersal tendency. The degree of nestedness measured at assemblage-level summarizes the response of species in the regional pool to these filters. Nestedness analysis alone, however, fails to elucidate processes that structure assemblage composition across a region. Approaches that utilize more stringent null models and examine the spatial response of nested and idiosyncratic taxa to ecological factors are essential if the processes that generate nested pattern are to be understood.
We would like to thank Alan Bedford and Nick Stewart for chironomid and charophyte identification, respectively, John Hopkins for help with plant identification, Jo Vosper, Ann Torr, Andy Hawley, Matt Frost and Richard Ticehurst for their combined help in the field and laboratory, Pierre Legendre for his advice on spatial autocorrelation analyses, Mark Lomolino for providing randomization software, Jeremy Clitherow and Ray Lawman for information and advice regarding sites, and English Nature for their financial support.