Functional trait-based approaches are increasingly used for studying the processes underlying community assembly. The relative influence of different assembly rules might depend on the spatial scale of analysis, the environmental context and the type of functional traits considered. By using a functional trait-based approach, we aim to disentangle the relative role of environmental filtering and interspecific competition on the structure of European ant communities according to the spatial scale and the type of trait considered. We used a large database on ant species composition that encompasses 361 ant communities distributed across the five biogeographic regions of Europe; these communities were composed of 155 ant species, which were characterized by 6 functional traits. We then analysed the relationship between functional divergence and co-occurrence between species pairs across different spatial scales (European, biogeographic region and local) and considering different types of traits (ecological tolerance and niche traits). Three different patterns emerged: negative, positive and non-significant regression coefficients suggest that environmental filtering, competition and neutrality are at work, respectively. We found that environmental filtering is important for structuring European ant communities at large spatial scales, particularly at the scale of Europe and most biogeographic regions. Competition could play a certain role at intermediate spatial scales where temperatures are more favourable for ant productivity (i.e. the Mediterranean region), while neutrality might be especially relevant in spatially discontinuous regions (i.e. the Alpine region). We found that no ecological mechanism (environmental filtering or competition) prevails at the local scale. The type of trait is especially important when looking for different assembly rules, and multi-trait grouping works well for traits associated with environmental responses (tolerance traits), but not for traits related to resource exploitation (niche traits). The spatial scale of analysis, the environmental context and the chosen traits merit special attention in trait-based analyses of community assembly mechanisms.

Data analyses

Different trait-based approaches have been used to distinguish the stochastic and deterministic (environmental vs. biotic filtering) processes that structure biotic communities. The approach we use can disentangle the role of environmental filtering and competitive exclusion by analysing the relationship between species pair co-occurrence and functional dissimilarity (2). From this analysis, three different patterns might emerge. First, if species with similar functional traits co-occur more often than expected by chance, the relationship between co-occurrence and functional dissimilarity of pairs of species will be significant and negative (i.e. environmental filtering process). Contrary to this, if species with divergent traits co-occur more often than expected at random, the relationship will be significant and positive (i.e. competitive exclusion process). Finally, non-significant relationships between co-occurrence and functional dissimilarity of species pairs are also possible (i.e. neutral theory processes). This would be the case where species co-occur independently of their functional similarity, or alternatively, if environmental filtering and competition exclusion are simultaneously at work with similar contributions. Here, we assume that two species co-occur when they occur spatially in the same community, although they might not share the same foraging time.

The co-occurrence index for each species pair was calculated within each species x site (European and regional scales) and species x bait (local scale) matrix. Data for the co-occurrence analyses consist in binary presence-absence matrices, where each row was a species, each column a site (or a bait), and the entries were presence (1) or absence (0) of a species in a site or a bait. Pairwise co-occurrence was calculated using the Jaccard index of similarity (JIab) for each pair of species in each matrix (47):

JIab=AB÷A+B+AB

where A and B are the number of sites where only species a and species b occur, respectively, and AB the number of sites where species a and b co-occur. The Jaccard similarity index takes values between 0 and 1, where 0 means that the two species are never found in the same site, and in our case, that co-occurrence is null; while 1 indicates that the two species are always together, and in our case, that the co-occurrence is total.

In order to measure functional dissimilarity between species pairs, we computed Gower’s dissimilarity between two species based on each functional trait separately, pooling traits according to whether they are ‘ecological tolerance’ or ‘ecological niche’ traits, and pooling all traits together. We used Gower’s dissimilarity, so that we would be able to deal with quantitative and qualitative traits (48). To compute it, we used a functional matrix where rows were species, columns were traits, and cell values were the trait values. Since Gower's dissimilarity depends on the number of species in the matrix, it was only calculated for each pair of species with data from the largest scale (Europe) where the number of species is highest. For each pair of species, nine functional dissimilarities were calculated: one with all functional traits together; one with only the ecological niche traits; one with the ecological tolerance traits; and one for each of the six traits separately. For these computations we used the ‘vegan’ (49) and ‘cluster’ (50) packages in R software v. 3.2.2 (51).

The relationship between the functional dissimilarity and the co-occurrence index between species pairs was tested by using linear models. Given the large number of zeros in the co-occurrence index and failure to meet the normal assumptions, we carried out the analyses in two steps. First, we transformed the co-occurrence index into a binary variable indicating whether or not there was occurrence of the pair of species in each matrix. We used a generalized linear model with a binomial distribution and a logit link function to perform the analysis (hereafter, binary co-occurrence analysis). In a second step, we applied a general linear model to make the model with the co-occurrence index where the pair of species occur at least once in the matrix (hereafter, co-occurrence strength analysis). In this case, the co-occurrence index was log-transformed to satisfy normality assumptions. We performed 18 analyses at the European scale (nine analyses for binary occurrence matrices and nine for co-occurrence strength matrices, these last nine comprising one analysis with all traits together, two analyses corresponding to each group of traits, and six analyses corresponding to each trait separately), 90 analyses at the biogeographic scale (forty-five for binary occurrence matrices and forty-five for occurrence strength matrices, of which nine analyses corresponded to each of the five biogeographic regions), and 333 analyses at the local scale (117 for binary occurrence matrices and 216 for co-occurrence strength matrices, comprising 37 analyses with all traits together, 37 for each group of traits and 222 for each singular trait). It is worth noting that binary co-occurrence analyses were only performed in locations where more than five pairs of species showed values of co-occurrence=0. Generalized and general linear models were conducted using the ‘stats’ package in R.