Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
Animal ecology could benefit from a well-defined trait-based framework, mostly applied in plant ecology, to further develop predictions of animal communities under various environmental conditions. We extended the functional approach to a multitrophic system by combining plant and ant traits in relation to environmental conditions to study the relationships between these three components.
We sampled plant and ant abundances along an aridity gradient in grazed and ungrazed conditions in the arid steppes of eastern Morocco. We measured five plant functional traits related to water stress and grazing resistance and six ant functional traits related to body size, dispersal and behaviour. We related each component (environment, vegetation and ants) using Mantel partial correlations to uncover the causal structure between components and using a fourth-corner analysis to describe the effects of the environment and vegetation on ant communities.
Results indicated that vegetation had a direct effect on ant community composition while the environment only had an indirect effect on ant community composition through vegetation structure. This result was consistent when looking at both the taxonomic and functional composition of communities, but correlations were stronger when based on taxonomic composition. Aridity was the variable most significantly linked with ant functional traits
Synthesis. The use of functional traits in animal ecology is relatively new, and an increase in trait-based community ecology studies that include more than one trophic level would be beneficial in identifying trait-based patterns in multitrophic communities. This new approach could become very useful in identifying mechanistic explanations of multitrophic community assembly and making predictions about their evolution under changing environmental conditions. It could also be of practical use in conservation biology in assessing habitat quality.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The current trend in plant community ecology is to rely on functional traits to explain species assemblages along environmental gradients. This approach relies heavily on the notion of trait-based environmental filtering (Keddy 1992) and on the measurement of relevant functional traits along environmental gradients (Grime 2001; McGill et al. 2006). It holds great potential for accurate and quantitative predictions of community composition. An interesting avenue in community ecology would be to extend the functional approach, developed mostly from plant ecology, to other trophic levels. Beyond the disciplinary differences between plant and animal ecology, a predictive functional approach would be useful in understanding how the environment is linked to the functional structure of ecological communities that are described at more than one trophic level. This approach could provide new insights into community assembly mechanisms and could also be relevant for biodiversity monitoring and ecosystem service assessment (Vandewalle et al. 2010; Luck et al. 2012).
Animal ecologists have a long tradition of grouping animals based on ecological, rather than on taxonomic, similarities (e.g. Root 1967). However, the newer ‘trait-based’ approach in plant ecology, which we use here, concentrates on the distribution of the traits themselves in the community rather than on nontaxonomic groupings that are based on suites of traits. This new notion of trait-based community assembly is less developed in animal ecology than in plant ecology. The few studies that have measured functional traits in animals were mainly interested in characterizing the environment–animal relationships and have been concerned mostly with birds (Vandewalle et al. 2010), benthic macroinvertebrates (Statzner, Dolédec & Hugueny 2004), beetles (Ribera et al. 2001; Moretti et al. 2010) and ants (Wiescher, Pearce-Duvet & Feener 2012). Some studies have looked at the functional relationships between plant traits and phytophagous insects (Peeters, Sanson & Read 2007) or butterflies (Dennis et al. 2004), but these studies did not include environmental gradients. A more global approach would be to consider environment–vegetation–animal relationships simultaneously using the functional approach. To our knowledge, this has only been done with grasshoppers by van der Plas, Anderson & Olff (2012) and with carabid beetles by Gobbi et al. (2010). Plant–insect systems might be a good place to start because insect communities are usually linked to plant communities (e.g. Siemann 1998) and because of the relative ease with which they can be sampled compared with vertebrate taxa. Insect species are also interesting because of their widespread use as indicators (Vandewalle et al. 2010). We chose ants because of their importance in arid ecosystems (Hölldobler & Wilson 1990) and because of their relevance as indicator species (Agosti et al. 2000; Vandewalle et al. 2010).
We investigated the application of functional traits to studying community structure along environmental gradients using two trophic levels, namely plants and ants. Recently, Frenette-Dussault et al. (2012) described the changes in the functional structure of the vegetation in the arid steppes of eastern Morocco along gradients of aridity and grazing exclosure. Here, we have extended the application of functional traits to ant communities in the same region. Specifically, we asked two questions: (i) What are the effects of the environment and vegetation structure on ant functional community assemblages in arid steppes? (ii) How are the environment, vegetation structure and ant communities causally related at the taxonomic and functional levels? With respect to this second question, we considered four plausible alternative hypotheses that differ in their causal assumptions (Fig. 1). We hypothesized a direct effect of both the environment and vegetation on ant functional traits due to the known influences of aridity (e.g. Pfeiffer, Chimedregzen & Ulykpan 2003) and vegetation composition (e.g. Siemann 1998) on ant communities.
Materials and methods
Study Sites and Abiotic Environmental Variables
We carried out this study in the arid steppes of eastern Morocco. The climate of this region has been described as Mediterranean arid, with cold winters and hot dry summers (Le Houérou 1995); average yearly minimum and maximum temperatures are about 10 and 25 °C, respectively, and annual precipitation ranges between 150 and 250 mm. These steppes, which are collectively owned, have become degraded due to overgrazing by sheep and goats, the harvesting of woody species and the absence of fallow periods. We selected six sites (see Appendix S1 in Supporting Information) along a short precipitation gradient, each having both grazed and ungrazed areas. At each site, we calculated the aridity index, defined as the ratio of potential annual evapotranspiration to total annual precipitation, and as described in Frenette-Dussault et al. (2012). The ungrazed areas in each site were permanent exclosures. The aridity index and duration of exclosure were not independent because sites differed in duration of exclosure. At each site, we sampled four plots: two in grazed and two in ungrazed conditions, except for the Enjil-2 site in which we sampled two ungrazed plots only, for a total of 22 plots. Distance between two plots was always >800 m.
Vegetation and Ant Sampling
In each plot, we measured plant and ant abundances. Detailed methodology of plant abundance measurement is included in Frenette-Dussault et al. (2012). Briefly, we assessed plant species abundance based on plant cover, and we measured the percentage of bare ground using the point-intercept method on 20 × 20 m2 plots. This was done at the peak of standing biomass (April–May 2010). Ant abundance was estimated at each plot with four pitfall traps, which were located at the corners of the vegetation plots next to a perennial plant (Hingrat et al. 2007), over a 3-week period from 31 May to 19 June 2010. This period corresponded to the highest above-ground ant activity (Hingrat et al. 2007). Pitfall traps consisted of a PVC pipe (17 cm height, 10 cm diameter) topped with a funnel placed flush to the ground. A small container was placed inside the PVC pipe just under the funnel and filled with 30 mL of ethylene glycol to collect ants. Pitfall traps were emptied biweekly and ants were brought back to the laboratory where they were identified to species (or morpho-species, in a few cases). Ant abundances of the four pitfall traps per plot were averaged to estimate abundance. Hence, the plot, and not the pitfall trap, was our sampling unit (n = 22).
In each plot, we measured five plant functional traits [specific leaf area (SLA), leaf dry matter content (LDMC), 13C isotope ratio (δ13C), succulence of leaves or stem, and onset of flowering] based on standardized protocols (Cornelissen et al. 2003) for species comprising at least 85% of total relative abundance (Pakeman & Quested 2007). We chose these traits because of their functional links with water stress and grazing. Additional methodological details on plant trait measurements are available in Frenette-Dussault et al. (2012).
We measured six ant functional traits that were related to their ecological niche: head length (HL), relative eye length (REL), relative leg length (RLL), feeding guild (granivorous, omnivorous or insectivorous), period of activity (diurnal or nocturnal) and colour (dark-coloured or not) (Ribera et al. 2001; Bihn, Gebauer & Brandl 2010). HL is well correlated with body mass (Kaspari & Weiser 1999) and other body parts (eye length, leg length, and thorax length and width; this study). We considered this trait relevant, because it estimates body size. Body size is related to many aspects of an individual's functioning in the environment, like resource consumption and metabolism (Peters 1983). REL and RLL correspond to eye and leg lengths divided by head length, respectively (Bihn, Gebauer & Brandl 2010). Eye size in arthropods is related to prey detection and visual recognition. We assumed that eye morphology could be associated with habitat type (Bauer et al. 1998). Leg size is related to the size–grain hypothesis (Kaspari & Weiser 1999), which stipulates that animals with longer legs will be able to move faster, but will be prevented to access small resource patches (e.g. small cracks in the soil), whereas animals with smaller legs will be slower, but will be able to access such small patches.
We used a binocular equipped with a micrometer (Leica Microsystems M80, Wetzlar, Germany) for the morphometric measurements. Before any measurement was carried out, the ant was laid out flat on an adhesive surface to facilitate manipulations. We measured HL as the maximum length between the clypeus and the occipital margin. We measured eye length as the maximum distance from one margin to the other when the head was flat on the adhesive surface. We measured leg length as the sum of tibia and femur length of the third leg (i.e. the hindleg). We considered only worker ants in the trait measurements. We measured those traits on a minimum of 10 workers per species. We did not consider rare species (i.e. with <20 individuals sampled). None of the species we sampled exhibited strong caste dimorphism. We verified intraspecific variability in head length for three abundant species (see Appendix S2). We did not find any consistent effect of aridity or grazing on intraspecific trait variability, so we considered a single trait value per species.
Causal Relationships between the Environment, Vegetation Structure and Ant Communities
We used simple and partial Mantel correlations to assess (partial) independence relationships between the environment, vegetation structure and ant communities. Partial Mantel correlations (Legendre & Legendre 1998) allow one to quantify the degree of correlation between two dissimilarity matrices conditional on a third one in the same way that a partial Pearson correlation coefficient quantifies the correlation of two variables conditional on a third one. We tested the plausibility of four alternative path models (Fig. 1) to explain such relationships based on these dissimilarity matrices. These alternative path models were tested using d-separation tests of path models (Shipley 2000a,b).
We first created dissimilarity matrices for the environmental variables, for vegetation structure and for ant communities. For the environmental dissimilarity matrix (ENV), we computed the Euclidean distance among plots based on two variables: aridity index and time-since-exclosure (0 for grazed sites). For the vegetation (VEG) and ant dissimilarity matrix (ANT), we used the Bray–Curtis index (Legendre & Legendre 1998) on the site by species abundance matrix. We applied a log10(x+1) transformation to the ant abundance matrix. This was necessary to reduce the influence of a single dominant species (Monomorium areniphilum Santschi) which occurred in all 22 plots, and that accounted for 75·8% of all collected individuals. We then used these dissimilarity matrices to compute simple and partial Mantel correlations.
To perform the same analyses at the functional level, we first computed community-weighted mean (CWM) trait matrices by multiplying the abundance matrices (the log10-transformed matrix in the case of ants) by the functional trait matrices for plant and ant data (Garnier et al. 2004), thereby giving the trait value per plot of the average individual irrespective of its taxonomic affiliation. We transformed those CWM trait matrices into plant (VEG) and ant (ANT) functional dissimilarity matrices using the Euclidean distance.
Ant Community Assembly based on Functional Traits
We used the fourth-corner analysis (Dray & Legendre 2008) to evaluate functional relationships between the environment and ant functional traits that might explain community assembly. This analysis allows one to test for the significance of all pairwise combinations of species traits and environmental variables by analysing three matrices simultaneously: sites by environmental variables, sites by species abundance data and species by functional traits. In this analysis, the ‘environment’ included the aridity index, the time-since-exclosure, the percentage of bare ground and the CWM trait variables of the vegetation. We used ‘model 2’ of Dray & Legendre (2008) to perform the permutations. This model preserves the original species assemblages and tests whether or not environmental characteristics affect the presence of such assemblages at particular sites. Again, we used the log10-transformed abundance data matrix to reduce the effect of M. areniphilum.
We ran all analyses within the R statistical environment (R Development Core Team, Vienna, Austria, 2011). We computed simple and partial Mantel correlations with the mantel and mantel.partial functions, respectively, from the vegan package (Oksanen et al. 2011). We performed the fourth-corner analysis with the fourthcorner function from the ade4 package (Dray & Dufour 2007). In all cases, we used 9999 permutations to assess significance.
Causal Relationships between the Environment, Vegetation Structure and Ant Communities
Although we had initially assumed that model 3 (Fig. 1) was the most plausible, this model was rejected because one of the predictions (ENV • ANT | VEG ≠ 0; see model predictions on Fig. 1) was not met. Instead, simple and partial Mantel correlations indicated that vegetation had a direct effect on ant community composition and environmental conditions had only an indirect effect on the ant communities through its effect on the vegetation (i.e. model 2 in Fig. 1). We obtained this result for both the taxonomic and functional composition analyses (Table 1). It is the only model where the data met all model predictions. Simple and partial correlations tended to be higher when based on taxonomic rather than functional composition. Also, the VEG-ANT correlation tended to be higher than the ENV-VEG correlation.
Table 1. Relationships between the environmental variables (ENV), vegetation structure (VEG), and ant community composition (ANT). The values indicate the Pearson correlation coefficients for simple (above the diagonal) and partial Mantel correlations (below the diagonal). Analyses were performed on a) plant and ant taxonomic composition and b) plant and ant functional trait composition. Ant abundance was log10-transformed. Significance was assessed with 9999 permutations
Aridity was the variable most related to ant functional traits with significant associations with almost all ant traits, except for head length and omnivory (Fourth-corner analysis, Table 2). Ant traits were negatively correlated with aridity, except for granivory and nocturnal activity. All associations between ant traits and time-since-exclosure were nonsignificant. Bare ground had less significant associations with ant traits than aridity, but responded similarly for relative leg length, granivory, insectivory and colour. Two plant traits (SLA and LDMC) were positively associated with one (relative eye length) and two (head length, relative leg length) ant traits, respectively. Omnivory was not associated with any of the environmental and vegetation variables. Insectivory decreased with aridity and tended to respond in the opposite direction to granivory for the significant associations with vegetation CWM traits. Diurnal activity and dark-coloured species decreased with aridity. Colour responded similarly to insectivory with respect to the environment and vegetation.
Table 2. Results of the fourth-corner analysis. This analysis indicates the tendency of the permutation results for each trait–environment combination. Pairwise relationships at the 5% significance level are indicated in dark grey for positive association, light grey for negative association and white for nonsignificant association. For each significant result, a Pearson correlation coefficient (r) for quantitative–quantitative trait associations or an ANOVA-like pseudo-F (F) for quantitative–qualitative trait associations is shown (Dray & Legendre 2008)
Causal Relationships between the Environment, Vegetation Structure and Ant Communities
Our results showed a direct link between vegetation and ant communities, but we did not find a direct environmental effect as we had hypothesized (Fig. 1, model 3). Instead, our results suggested that ant community composition was only indirectly affected by environmental conditions through its effect on vegetation (Fig. 1, model 2). It was somewhat surprising not to find a direct effect of environmental conditions on ant community composition. However, vegetation structure can be relatively more important than large-scale environmental factors like fire (Vasconcelos et al. 2008) or grazing (Calcaterra et al. 2010) in explaining ant community composition because these factors directly influence vegetation structure, which then regulates microclimatic conditions. Changes in microclimate, in turn, can directly affect ant activity (Kaspari 1993). Aridity, as we measured it, might act as a coarser filter, which excludes all ant taxa that are not already adapted. Environmental constraints on those ant taxa that have successfully passed through this regional filter are possibly more relevant at smaller spatial scales and mediated by local vegetation.
Although water stress is a major constraint on desert ants (Hood & Tschinkel 1990), ants of arid environments can be active over a broad range of environmental conditions (Bestelmeyer 1997) and can forage near their physiological tolerance due to their thermophilic nature (Hölldobler & Wilson 1990). For example, ants of the genus Cataglyphis, which were abundant in all of our study sites, are hot climate specialists (sensu Andersen 1995) that can forage when ground surface temperatures are above 60 °C (Hölldobler & Wilson 1990). A high physiological tolerance combined with a broad distribution thus limits the expression of direct environmental filtering. If the local environmental gradients are not sufficiently contrasting, then community average trait–environment relationships may not be able to identify a strong environmental filtering effect (e.g. Sonnier, Shipley & Navas 2010). We suggest that a combination of both taxonomic and functional approaches may be more suitable to making more accurate predictions of community composition.
Ant Community Assembly Based on Functional Traits
The fourth-corner analysis revealed interesting significant trait–environment associations. For instance, an increase in aridity and bare ground was associated with an increase in the abundance of granivorous species (Table 2). This result was consistent with the high presence of seed harvesting Messor sp. in the more arid sites. Insectivory, on the other hand, was negatively associated with aridity. This could be explained by a decrease in vegetation cover, which in turn could negatively affect the presence and diversity of phytophagous insects that are consumed by insectivorous ants (Lightfoot & Whitford 1990). Also, we would have expected a positive correlation between relative leg length and bare ground based on the size–grain hypothesis (Kaspari & Weiser 1999), but we found a rather weak negative correlation. Again, this probably reflects the ‘shortness’ of our aridity gradient. Grazing by sheep was not correlated with any of the ant traits. It is possible that the duration of the exclosures (from 2 to 14 years) and the resistance of ants to the direct effect of trampling during grazing (Heske & Campbell 1991) may explain the absence of a correlation.
We found significant associations between plant and ant traits (Table 2). As 13C isotope ratio, succulence of leaves and onset of flowering were correlated with aridity (Frenette-Dussault et al. 2012), it would also explain why ant traits showed similar responses to these plant traits than to aridity. Other significant associations were more difficult to interpret. For example, it is not clear why plant communities with higher LDMC would favour bigger ants (Table 2). Perhaps, as LDMC was positively correlated with plant size and seed mass at the community level (Frenette-Dussault et al. 2012), bigger ants would be more apt to bring back bigger seeds to the nest. At this stage, it is not possible to say whether there is a real causal connection between these two functional traits or it is only a spurious correlation. Further trait-based studies of multitrophic community assembly will help to clarify whether such patterns are general or site specific.
Our approach was based on the measurement of functional traits to investigate trait-based environmental filtering, but we did not consider biotic interactions among ant species. Many studies of ant community assembly also pointed out the importance of interference competition among ant species as a strong determinant of community composition (Hölldobler & Wilson 1990). We rejected this hypothesis based on an environmentally constrained null model analysis (Peres-Neto, Olden & Jackson 2001). The results suggested that segregated species patterns (i.e. fewer species co-occurrences than would be expected by chance) were probably due to different environmental affinities among ant species (and a ‘short’ gradient) rather than to competition (see Appendix S3).
The application of the functional approach to animal ecology is still in its infancy, and we would strongly benefit from a theory similar to Grime's (2001) C-S-R plant strategies for animal taxa. Andersen (1995, 1997) proposed a functional classification of ant communities based on ant behavioural dominance and their responses to stress and disturbance. This classification has been compared to Grime's (2001) C-S-R theory, but the former has been mostly developed for Australian ants and is based on behavioural observations and species distributions. Although we mostly concentrated on functional traits that were related to morphology, it might be important to include behavioural traits as well (Andersen 1995, 1997). The next step in developing this functional classification would include the measurement of more varied functional traits and quantifying the relationships between these traits and a diverse array of environmental conditions and vegetation structures. By focusing on functional environment–vegetation–ant trait relationships, we might be able to identify recurring patterns in multitrophic community ecology and provide explanations as to how such communities are assembled that are not limited by taxonomy. An even more integrative approach would be to consider the Universal Adaptive Strategy Theory (Grime & Pierce 2012), which strives to identify a common classification for disparate taxa by placing the adaptive strategies of plants, animals and micro-organisms within the same theoretical framework. Quantifying the proportion of carbon and nitrogen invested in traits involved in resource acquisition, metabolic maintenance or regeneration could be a determinant factor in developing such a classification (Grime & Pierce 2012).
Finally, our results have potential future consequences to conservation biology. Ants are a functionally important group in arid ecosystems due to their numerical abundance and their ecological effects on various ecological processes, such as seed dispersal (Hölldobler & Wilson 1990). In such ecosystems, they can also represent a high proportion of vertebrate diets (e.g. Bourass et al. 2012). Ants are widely used as indicator species in biomonitoring studies due to their sensitivity to environmental changes (Agosti et al. 2000; Vandewalle et al. 2010). A conservation approach that would consider biomass, together with taxonomic and functional relationships between animals and plants along environmental gradients, would be very useful for characterizing habitats, assessing their quality (in terms of resource availability), providing a mechanistic explanation of multitrophic community assembly, and making predictions about their evolution under climate change scenarios (Dennis 2010). Conservation plans that integrate all these aspects could potentially identify the most important habitats or areas to preserve. This is utterly relevant in a world where limited economic resources are devoted to conservation.
Funding and supervision of this study were provided by the Emirates Center for Wildlife Propagation (ECWP) under the leadership of the International Fund for Houbara Conservation (IFHC). We are grateful to H.H. Sheikh Mohamed bin Zayed Al Nahyan, Crown Prince of Abu Dhabi and Chairman of the IFHC, and H.E. Mohammed Al Bowardi, Deputy Chairman of IFHC, for their support. We thank J.F. Léger and S. Boullenger for their supervision and guidance and Dr. Simon Pierce and an anonymous reviewer for thoughtful comments that improved the manuscript. We also thank F.O. Ezza, A.S.R. Sanz, M. Bidat and H. Hdidou for their assistance during field work, M. Sbai, H. Saada and S. Gyssels for laboratory work, and A. François and C. Galkowski for ant identification. We are grateful to the Haut Commissariat aux Eaux et Forêts et à la Lutte Contre la Désertification (Morocco) for authorizing access to Tirnest and Enjil-2 sites.