Oecologia Does morphology predict trophic position and habitat use of ant species and assemblages? --Manuscript Draft-- Manuscript Number: OECO-D-14-00562R1 Full Title: Does morphology predict trophic position and habitat use of ant species and assemblages? Article Type: Community ecology – original research Corresponding Author: Heloise Gibb CSIRO Canberra, Australian Capital Territory AUSTRALIA Order of Authors: Heloise Gibb Jakub Stoklosa David Warton Alexandra Brown Nigel Andrew Saul Cunningham Response to Reviewers: See attached letter for the response to referees Abstract: A functional traits-based theory of organismal communities is critical for understanding the principles underlying community assembly, and predicting responses to environmental change. This is particularly true for terrestrial arthropods, of which only 20% are described. Using epigaeic ant assemblages, we asked: 1) Can we use morphological variation among species to predict trophic position or preferred microhabitat?; 2) Does the strength of morphological associations suggest recent trait divergence?; 3) Do environmental variables at site scale predict trait sets for whole assemblages. We pitfall-trapped ants from a revegetation chronosequence and measured their morphology, trophic position (using C:N stoichiometry and isotope ratios) and characteristics of microhabitat and macrohabitat. We found strong associations between high trophic position (low C:N and high δ15N) in body tissue and morphological traits: predators were larger, had more laterally-positioned eyes, more physical protection and tended to be monomorphic. In addition, morphological traits were associated with certain microhabitat features, e.g., smaller heads were associated with the bare ground microhabitat. Trait-microhabitat relationships were more pronounced when phylogenetic adjustments were used, indicating a strong influence of recent trait divergences. At the assemblage level, our 4th corner analysis revealed associations between the prevalence of traits and macrohabitat, although these associations were not the same as those based on microhabitat associations. This study shows direct links between species-level traits and both diet and habitat preference. Trait-based prediction of ecological roles and community structure is thus achievable when integrating stoichiometry, morphology and phylogeny, but scale is an important consideration in such predictions. Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation 1 Does morphology predict trophic position and habitat use of ant species and assemblages? Gibb, H. 1,2*, Stoklosa, J. 3, Warton, D.I.3, Brown, A.M. 3,4, Andrew, N.R. 5, Cunningham, S.A. 2 1 Department of Zoology, La Trobe University, Melbourne, VIC 3068, Australia 2 CSIRO Ecosystem Sciences, Black Mountain, ACT 2601, Australia 3 School of Mathematics and Statistics and Evolution & Ecology Research Centre, The University of New South Wales, NSW 2052, Australia 4 Evolution & Ecology Research Centre and School of Biological, Earth and Environmental Sciences, The University of New South Wales, NSW 2052, Australia 5 Centre for Behavioural and Physiological Ecology, Discipline of Zoology, School of Environmental and Rural Science, University of New England, NSW 2351, Australia Corresponding author: email: h.gibb@latrobe.edu.au; phone: +613 9479 2278 Manuscript Click here to download Manuscript: Gibb et al. Oecologia manuscript - revision.docx Click here to view linked References 2 Abstract 1 A functional traits-based theory of organismal communities is critical for understanding the 2 principles underlying community assembly, and predicting responses to environmental 3 change. This is particularly true for terrestrial arthropods, of which only 20% are described. 4 Using epigaeic ant assemblages, we asked: 1) Can we use morphological variation among 5 species to predict trophic position or preferred microhabitat?; 2) Does the strength of 6 morphological associations suggest recent trait divergence?; 3) Do environmental variables at 7 site scale predict trait sets for whole assemblages. We pitfall-trapped ants from a 8 revegetation chronosequence and measured their morphology, trophic position (using C:N 9 stoichiometry and isotope ratios) and characteristics of microhabitat and macrohabitat. We 10 found strong associations between high trophic position (low C:N and high δ15N) in body 11 tissue and morphological traits: predators were larger, had more laterally-positioned eyes, 12 more physical protection and tended to be monomorphic. In addition, morphological traits 13 were associated with certain microhabitat features, e.g., smaller heads were associated with 14 the bare ground microhabitat. Trait-microhabitat relationships were more pronounced when 15 phylogenetic adjustments were used, indicating a strong influence of recent trait divergences. 16 At the assemblage level, our 4th corner analysis revealed associations between the prevalence 17 of traits and macrohabitat, although these associations were not the same as those based on 18 microhabitat associations. This study shows direct links between species-level traits and both 19 diet and habitat preference. Trait-based prediction of ecological roles and community 20 structure is thus achievable when integrating stoichiometry, morphology and phylogeny, but 21 scale is an important consideration in such predictions. 22 Key words: Carbon to Nitrogen ratio; delta N; microhabitat; morphological traits; 23 stoichiometry 24 3 Introduction 25 Trait-based approaches are becoming increasingly important in understanding community 26 structure and species responses to global change (McGill et al. 2006; Westoby and Wright 27 2006; Bihn et al. 2010; Silva and Brandão 2010; Gibb and Parr 2013). Although phylogeny 28 is an indicator of function, its value can be limited because of ecological divergences in some 29 clades, and convergences among others. In contrast, the use of species traits allows us to 30 examine commonalities across communities that differ in their evolutionary history, but 31 which often include ecologically similar species performing similar roles. While the use of 32 traits in understanding the ecology of plants is relatively well developed (Westoby and 33 Wright 2006; Ackerly and Cornwell 2007; Cornwell et al. 2008), this approach is in its 34 infancy in animal ecology. The ecology of many vertebrate species is well known, making 35 predictive models of responses of individual species possible for many groups. However, our 36 limited understanding of how species traits interact with the environment is particularly 37 problematic for invertebrates because these taxa are poorly studied at the species level. 38 Considerable interest in traits has resulted in the development of new statistical techniques 39 that allow us to examine how species traits relate to the environment (e.g., Shipley et al. 40 2006; Brown et al. 2014; Stoklosa et al. 2014), and these techniques have the potential to 41 rapidly advance our understanding of invertebrate trait-environment interactions. 42 A range of trait types may be considered useful for predicting the ecology of a species. To 43 address questions related to global change, features of organisms that affect their likelihood 44 of persistence in different environments are an obvious target. These include morphological, 45 physiological, behavioural or life history traits that interact with structural elements of the 46 environment (Sarty et al. 2006; Bihn et al. 2010; Wiescher et al. 2012; Gibb and Parr 2013), 47 disturbance tolerance (e.g., Arnan et al. 2013; Yates et al. 2014) or climate (Chown 2012; 48 4 Diamond et al. 2012; Wiescher et al. 2012) or that favour success in the face of species 49 invasions (Funk et al. 2008). Body size is probably the best-studied trait of animals, with 50 evidence that it responds strongly to habitat disturbance and climate, both within and among 51 species (Chown and Gaston 2010; Entling et al. 2010). However, a range of other, less well 52 understood morphological traits are likely to be important in determining species responses to 53 global change (e.g., eye position, Gibb and Parr 2013). Although many of these 54 morphological traits have hypothesised functions (e.g., Table 1), there is usually little 55 empirical testing of those functions. 56 Elemental stoichiometry is increasingly recognised as a means to determine the diet of 57 species, another important trait. Predators have higher percentages of nitrogen in their 58 tissues, possibly as a result of higher N in the predator diet, a requirement for more N-rich 59 body parts (e.g., muscles, protein-rich cuticle) or to support N-based chemical weaponry 60 (Fagan et al. 2002; Davidson 2005). A lower C:N ratio in body tissue is therefore interpreted 61 as indicating a more predatory feeding profile. In addition, stable isotopes of carbon and 62 nitrogen are informative because enrichment of the heavier isotopes of nitrogen and carbon 63 occurs with each trophic level (DeNiro and Epstein 1979; DeNiro and Epstein 1981). The 64 trophic position of species may therefore be characterised by chemical analyses of their 65 tissues, which is a considerable improvement on previous observational approaches to 66 understanding diet. There is also potential to link these continuous measures of trophic 67 position with easily recorded traits, such as morphological traits. 68 Traits are, of course, also a product of phylogenetic history, such that more closely related 69 species will, on average, have more similar traits (Felsenstein 1985; Harvey and Pagel 1991). 70 For this reason many evolutionary studies use phylogenetic adjustments of the trait analysis 71 to account for this non-independence of species (Freckleton 2009). In this study, we chose to 72 examine correlations with trait values both with and without phylogenetic correction. This 73 5 allows the possibility of detecting patterns in one or the other model, or both. If a trait 74 correlation with an environmental variable is strongest without phylogenetic adjustment, this 75 suggests trait divergences deeper in the phylogeny dominate, with more recent divergences 76 not supporting the association. If a trait correlation is strongest with phylogenetic adjustment, 77 then recent trait divergences are strongly contributing to the pattern, but the deeper 78 phylogenetic divergences constrain the pattern. With evolutionary processes creating the 79 underlying pattern of trait diversity, species assemblages are “filtered” by the abiotic 80 environment and the milieu of species interactions is important in determining which species 81 and therefore which traits persist or dominate in an assemblage (McGill et al. 2006; Wiescher 82 et al. 2012). 83 In this study, we examine how traits change along a biotic gradient created through a 84 revegetation chronosequence, where sites differ in the time since replanting, which 85 corresponds to differences in a range of habitat structural measures. We address the 86 following questions: 1) Can we use morphological variation among species to predict: a) 87 trophic position; or b) preferred microhabitat?; 2) Does the strength of these morphological 88 associations suggest recent trait divergence?; 3) Do environmental variables at site scale 89 predict trait sets for whole assemblages of ants? 90 Material and Methods 91 Study sites description 92 The study was conducted in south eastern Australia, at 30 sites within 150 km of the city of 93 Canberra (35°15’S 149°08’E), at altitudes 450 to 720 m above sea level (Fig. S1). The region 94 experiences a relatively dry continental climate with warm to hot summers and cool to mild 95 winters. The vast majority of native vegetation in the study area has been cleared for grazing, 96 crops or urban development and the remaining vegetation is mainly open Eucalyptus 97 6 woodlands. Government-assisted revegetation has been conducted on many properties in the 98 study region since the early 1980s. Two methods were used: tube stock, where nursery-99 raised saplings were planted individually at distances between 2 and 5 m apart; and direct-100 seeding, where seeds were planted into ploughed furrows at higher densities. Both planting 101 types were dominated by a mix of Acacia and Eucalyptus species. 102 To assess the relationship between ant morphological traits and changes in habitat 103 characteristics across a successional gradient, five sites belonging to each of the following six 104 categories were selected : 1) pastures without trees (“pastures”); 2) revegetated pastures 105 planted with tube stock between 1998 and 2001 (“young tube stock”); 3) revegetated pastures 106 planted by direct-seeding between 1998 and 2001 (“young direct-seeded”); 4) revegetated 107 pastures planted with tube stock between 1989 and 1994 (“old tube stock”); 5) revegetated 108 pastures direct-seeded between 1989 and 1994 (“old direct-seeded”); and 6) remnant areas 109 (“remnants”) that had never been cleared and had been protected from heavy livestock 110 grazing for a minimum of 10 years (Fig. S1). One of the remnant sites experienced 111 occasional light grazing from cattle and evidence of kangaroos and rabbits was present in all 112 sites. Remnant and revegetation patches were of similar sizes and sites of different categories 113 were non-adjacent and spatially interspersed, but pastures were unavoidably part of a 114 continuous matrix of agricultural land. Landform was distributed evenly amongst site 115 categories. Although sites were selected because they fitted these specific categories, the 116 variation in habitat structure variables among sites was relatively continuous. 117 Pitfall sampling 118 Epigaeic ants were sampled using pitfall traps, which were open for two eighteen-day sessions in each 119 of the 30 sites in November 2007 and April 2008. Traps were left in place for a week before opening, 120 in order to avoid digging in effects (Greenslade 1973). Traps were 7 cm in diameter and 8 cm in 121 7 depth and were filled with 100 ml of propylene glycol. They were placed in two 2 x 2 m squares, with 122 traps at each corner and one in the centre (n = 5 traps) (Fig. S1). These trap squares were placed 123 halfway along the long edge of each site, with the first set of traps 20 m from the edge and the 124 following set 50 m from the edge of the site. Pitfall traps were protected from direct rainfall by a 125 black plastic plate (15 cm diameter), supported approximately 5 cm above ground level by pegs. 126 All specimens were identified to genus using Shattuck (1999). Morphospecies were verified, and 127 some genera were identified to species level, by Dr Steve Shattuck (Australian National Insect 128 Collection, CSIRO, Canberra, Australia). Assemblage analyses were conducted using ant 129 ‘occurrences’, i.e., the number of traps per site in which a species occurred. 130 Micro- and macrohabitat characteristics 131 Habitat surveys were conducted in November 2007 at two scales: in association with pitfall traps and 132 in 20 m x 20 m plots. Data associated with pitfall traps were used to characterise the microhabitats 133 used by collected species. For microhabitat data, the percentage of bare ground and leaf litter and 134 percentage canopy cover within a 1 m diameter circle, centred on each pitfall trap, was recorded. 135 Plot-scale data were used to characterise the thirty sites and were used in analyses at the assemblage 136 level and we refer to this as macrohabitat. Two quadrats of 20 m x 20 m surrounding each of the two 137 pitfall trap groups were marked at each site. The number of Eucalyptus, Acacia and other trees and 138 shrubs in three height categories (0.5-2 m, 2-10 m and > 10 m) and diameter at breast height (DBH) 139 and basal diameter (at 10 cm) for up to 10 trees in each of these categories was recorded. Trees were 140 categorised as remnant, regrowth or planted in the revegetation process and aspect and slope recorded. 141 In each quadrat, characteristics related to coarse woody debris (CWD) were noted. Whereas the usual 142 definition for CWD includes debris with a minimum diameter of ≥ 10 cm (e.g., Gibb et al. 2005), our 143 definition included debris of ≥ 5 cm diameter, in order to include fallen Acacia stems that commonly 144 have smaller diameters. For lying CWD, we recorded maximum and minimum diameter, length, 145 8 percent contact with the ground and decay state. For standing CWD, basal diameter, height and decay 146 state were recorded. Decay state was classed into four categories: 1) outline of CWD intact and bark 147 present; 2) outline of CWD intact, but bark lost from 50% of the surface; some bleaching; 3) Bark 148 absent, outline of CWD less distinct; usually bleached; 4) CWD without clear outline; usually 149 bleached and broken. 150 At the corners and centre of each of the 20 x 20 m quadrats, smaller, 1 m x 1 m quadrats were placed 151 in which ground cover variables were recorded (a total of 5 replicates in each larger quadrat). In the 152 smaller quadrats, the percentage cover of leaf litter, grass litter, bare ground, grass, moss and dead 153 wood was estimated. In addition, the height of the tallest grass, the depth of leaf and grass litter and 154 the presence and taxon of origin of animal dung were recorded. 155 Sampling for stoichiometry and stable isotope analyses 156 In March 2008, live ants were hand-collected from twelve study sites (4 pastures, 4 young 157 tube stock and 4 remnant sites) to conduct chemical analysis, uncontaminated by pitfall trap 158 preservatives. A species list is provided in Appendix I. All specimens were collected 159 between 9 am and 6 pm. Our goal was to collect a representative sample of the diurnally-160 active ants in each site. Up to 30 workers of 8-17 ant species were collected by hand or 161 aspirator at each site. For very small ant species, such as Tapinoma spp., 30 individuals were 162 required to obtain sufficient mass for a single analysable sample. Smaller numbers were 163 collected for larger ant species but samples always consisted of at least two individuals. 164 Means for each species collected were calculated, based on a sample size of 2-3 (each sample 165 contained 2-30 ants, depending on their size). Samples were stored on dry ice, then in a -166 20ºC freezer prior to analysis. Abdomens were removed before analysis to avoid 167 complications related to gut contents (Blüthgen et al. 2003; Davidson 2005; Tillberg et al. 168 2006). 169 9 Soil samples were collected in April 2008 at all sites to use as a baseline for isotope values (Gibb and 170 Cunningham 2011). Grass, leaf litter and the top centimetre of soil were removed and a 10 cm deep 171 plug of soil (diameter 43 mm) was collected using an auger. Three samples were collected from a 172 transect at each site, with 6 m spacing between samples. Soils were stored in a -20ºC freezer, and 173 dried at 40ºC for 48 hrs prior to analysis. 174 Samples were analysed for total C and N content and for the ratio of the heavy to light isotopes of C 175 and N. The analysis was performed using a Sercon Hydra 20-20 isotope ratio mass spectrometer 176 (made in Crewe, UK) with an ANCA (automated nitrogen and carbon analyser) preparation system at 177 CSIRO Black Mountain Laboratories (Canberra, Australia). Samples were dried in an oven at 60ºC, 178 then approximately 1.0 mg of each sample was weighed into a tin capsule. Test and reference samples 179 were used to correct for any drift or carry-over in the instrument. References were calibrated for total 180 C% and N%. 15N/14N was measured relative to atmospheric N, while 13C/12C was measured relative to 181 Vienna Pee Dee Belemnite. Sample ratios were compared to this element-specific standard and 182 reported as δX, where δX = [(Rsample/Rstandard)-1)] x 1 000. Rsample and Rstandard are the ratio of heavy to 183 light isotopes for the sample and standard, respectively. δ values are reported as “per mil” deviation 184 from standards for the 13C/12C and 15N/14N isotopic ratios of soil, herbivores, ants and predators. 185 The soil-corrected δ15N for all individuals (ant δ15N – soil δ15N at the same site) was 186 determined in order to eliminate differences in ant δ15N resulting from site differences. The 187 mean of the soil-corrected δ15N across all samples within each species was used in analyses. 188 Although the trophic position of ant genera may shift across restoration gradients (Gibb and 189 Cunningham 2011), the relative position of species is generally constant (Ponsard and Arditi 190 2000; Gibb and Cunningham 2011), so it is appropriate to use a mean value across all sites in 191 which a species was collected. 192 10 Morphological traits 193 Seven continuous morphometric measures and four ordinal measures were selected to 194 describe species traits. Traits were selected because they had shown ecologically relevant 195 associations in previous studies or because we had distinct hypotheses regarding their 196 function (Table 1). Although colony size is considered a key measure of body size for ants, 197 this trait was not available for species in this study and our emphasis was on traits of workers, 198 rather than of colonies. Traits were measured on three individual workers of each species, 199 which had been stored in ethanol. When species were morphologically dimorphic (Pheidole 200 spp.) or clearly polymorphic (Amblyopone spp., Camponotus spp., Melophorus spp. and 201 Notoncus spp.), only minors were used, although polymorphism was included as an ordinal 202 trait (Table 1). For each ant, standard linear measurements were taken using an ocular 203 micrometer mounted on a dissecting microscope, accurate to 0.01 mm. The positioning of 204 species in trait-space using principal component analysis is shown in Appendix II. 205 Data analysis 206 For all analyses, habitat data was averaged across seasons because preliminary analyses 207 showed that seasonal variation was low. For each site we pooled ant abundances across 208 sampling events because composition remained similar between seasons. 209 Can we use morphological variation among species to predict trophic position or preferred 210 microhabitat and does the strength of morphological associations suggest recent trait 211 divergence? 212 For regression analyses, all continuous measures were log-transformed and regressed against 213 loge (Weber’s length) to obtain residuals for analysis that did not depend on body size. Data 214 for both sampling dates were pooled as they were significantly correlated (Environmental 215 data: Pearson’s correlation: r2 > 0.6, p < 0.001; Species data: Mantel test: p = 0.001, Rho = 216 11 0.67). We used mean values per species for trait and isotope measures and mean measures of 217 environmental variables across all locations at which a species occurred. 218 Consistent with previous studies (Blackburn and Gaston 1998; Garland et al. 1999), both 219 phylogenetic and non-phylogenetic regression approaches were used to examine the 220 relationship between species traits and predictors related to diet (n = 30 species) and 221 microhabitat (n = 37 species). Non-phylogenetic regression analyses were conducted using 222 the “lm” function on R (R Development Core Team 2013). Using phylogenetic independent 223 contrasts (PICs) (Felsenstein 1985; Garland and Ives 2000) in the package ape (Paradis et al. 224 2004) on R allowed us to determine if present day patterns in trait associations are a result of 225 a small number of large divergences across trait values in the evolutionary history of the 226 species examined. An ant phylogeny derived from Moreau et al. (2006) was used to perform 227 phylogenetic independent contrasts between morphological traits and microhabitat and diet 228 variables. The phylogeny included only one or two examples from each genus, so species 229 within genera were included as soft polytomies. Polytomies were resolved using the multi2di 230 function on ‘ape’ to allow us to run phylogenetic independent contrasts and phylogenetic 231 regressions. 232 If relationships were not phylogenetically independent, this did not necessarily mean that the 233 trait-environment relationship was not meaningful. However, it did mean the relationship 234 was more difficult to interpret as it may suggest either: 1) that environmental filtering of 235 species by traits has resulted in related (and morphologically similar) species occurring in 236 similar habitats or: 2) that the pattern is driven by other traits shared among related species. 237 Do environmental variables at site scale predict trait sets for whole assemblages of ants? 238 A new “4th corner” analysis “trait.mod” on R (R Development Core Team 2013) was used to 239 determine the relationship between traits and the environment (Stoklosa et al. 2014), at the 240 12 assemblage level. While analyses of associations between species traits and microhabitat 241 described above used variables measured at each pitfall trap and related them to individual 242 species, the fourth corner analysis related traits to site-level environmental variables using 243 data on assemblage composition, which includes abundance data. The fourth-corner analysis 244 considers conditional effects after accounting for all species abundances in the model, rather 245 than examining species one-at-a-time. A fourth-corner problem can be thought of as one that 246 involves using three tables describing environmental data (R), species abundances (L) and 247 species traits (Q) to determine how species traits relate to the environment (D) (Brown et al. 248 2014). The analysis fits a predictive model for species abundances as a function of 249 environmental variables, species traits and their interactions (Brown et al. 2014). This 250 model-based approach uses generalized estimating equations (GEE) (Liang and Zeger 1986) 251 to account for correlation between species observed at sites. Forward selection and the score 252 information criterion (SIC) (Stoklosa et al. 2014) with a BIC-type penalty was used to select 253 the most important environment-trait interactions. For this analysis, we used negative 254 binomial regression because responses were count-based and a LASSO penalty, which has 255 been shown to endow high predictive performance in species distribution models (Renner and 256 Warton 2013). 257 Results 258 Can we use morphological variation among species to predict: a) trophic position; or b) 259 preferred microhabitat? 260 Both standard and phylogenetic regressions revealed negative relationships between C:N of 261 body tissue and Weber’s length and sculpturing and positive relationships between C:N of 262 body tissue and eye width and eye position (Table 2, Fig. 1). In addition, standard 263 regressions showed a negative relationship between C:N and head width and mandible length 264 13 and width. For phylogenetic regressions with δ15N, we detected positive relationships for eye 265 position and pilosity and a negative relationship for polymorphism. Standard regressions 266 revealed a significant positive relationship with mandible length. Although C:N and δ15N 267 might be expected to be negatively correlated if both are indicative of the same measure of 268 trophic position, this relationship was not significant (F(1,28) = 0.15 , p = 0.702). 269 For microhabitat characteristics, phylogenetic regressions showed that bare ground was 270 negatively related to head width, mandible length and width and spinosity, while canopy 271 cover was negatively related to eye position and positively related to spinosity (Table 2, Fig. 272 2). Standard regressions revealed only a negative relationship between canopy cover and 273 sculpturing. 274 Does the strength of morphological associations suggest recent trait divergence? 275 Similar patterns were observed for adjusted and unadjusted regressions on morphological 276 traits and diet, suggesting that recent divergences were likely to be as important as deeper 277 divergences in these relationships (Table 2). Surprisingly, accounting for phylogeny resulted 278 in greater detection of relationships between traits and microhabitat than regression analyses 279 on unadjusted data. This suggests that relatively recent evolutionary divergences drove these 280 patterns and that they were obscured if more distant evolutionary divergences were not 281 accounted for (Table 2). 282 Do environmental variables at site scale predict trait sets for whole assemblages of ants? 283 The fourth corner analysis revealed a range of significant interactions between ant 284 assemblage traits and the environment (Figs 3, 4). These relationships were not always the 285 same as those detected at the species level because the fourth corner analysis considered the 286 whole assemblage of ants. Weber’s length was positively correlated with bare ground and 287 negatively correlated with shrub cover, suggesting that larger ants dominated more in more 288 14 open plantings. It was also positively related to slope. Pilosity was positively related to bare 289 ground, suggesting that hairier ants were more prevalent in more open habitats. 290 Polymorphism was negatively related to bare ground and canopy cover, indicating that 291 monomorphic ants were more prevalent where the ground layer was more complex and the 292 canopy sparser. Ants with narrow heads were most abundant where shrub cover and CWD 293 availability were highest. 294 Discussion 295 We used ant assemblages collected along an environmental gradient to determine if 296 morphological variation among species predicts trophic position or preferred microhabitat, 297 whether these relationships suggest recent trait divergence and if environmental variables at 298 site scale predict trait sets for whole assemblages of ants. Morphology did predict some of 299 the ecological differentiation in both trophic level and microhabitat use. Although analyses 300 accounting for phylogeny are often considered more conservative, the phylogenetically 301 adjusted approach allowed us to detect patterns that were obscured in the standard regression 302 approach. This suggests that phylogeny concealed, rather than confounded, some patterns. 303 At larger scales, the prevalence of traits in the assemblage was dependent on a range of 304 habitat variables, but these were not always the same as those determining the traits of 305 species at the local scale. This highlights not only the differences between species-based and 306 assemblage-based assessments of traits, but also the potential for some trait-environment 307 relationships to be scale dependent. Although the exact habitat elements measured changed 308 with scale, the scale dependence we see here may suggest that the important biotic filters are 309 different at different spatial scales. This could reflect the difference between scales of daily 310 activity and the scale of habitat selection or dispersal. In addition, the inclusion of abundance 311 in assemblage-level metrics means that species interactions contribute to trait-environment 312 15 relationships. While it makes sense that the outcomes could be scale dependent, it also serves 313 as a warning regarding how far one might be able to generalise predictions based on traits. 314 Does the strength of morphological associations suggest recent trait divergence? 315 Relationships between morphological traits and diet showed similar patterns, with or without 316 adjustments for phylogeny. In contrast, trait-microhabitat relationships were more 317 pronounced when phylogeny was accounted for. This suggests that recent and more distant 318 evolutionary divergences were of similar importance in determining relationships between 319 diet and morphology, but that recent divergences were of key importance for microhabitat-320 morphology relationships. Diet may be largely conserved phylogenetically amongst ants, 321 e.g., all ectatommines are considered to be predators. Traits related to diet might therefore be 322 more evolutionarily constrained than those related to microhabitat use. 323 Can we use morphological variation among species to predict trophic position? 324 Several traits were correlated with Carbon to Nitrogen ratio (C:N) and/or δ15N of body tissue, 325 both with and without phylogenetic adjustment. A lower C:N or higher δ15N indicates a more 326 predatory genus (DeNiro and Epstein 1981; Fagan et al. 2002; Davidson 2005). C:N 327 predators were larger than predominantly omnivorous ant species. C:N predators included 328 Myrmecia, Rhytidoponera, Tetramorium and Amblyopone, all of which are considered highly 329 predatory and most of which also had high δ15N values in a previous study (Gibb and 330 Cunningham 2013). Most of these genera include relatively large species, in contrast with 331 previous findings that predatory ants are smaller on average in the New World (Weiser and 332 Kaspari 2006). While overall size was greater in more predatory species, analysis of δ15N 333 showed that variation in the size and physical proportions of workers within a colony (caste 334 polymorphism) was lower. This might be expected if task specialisation was lower in species 335 that are predominantly predatory (Oster and Wilson 1978). 336 16 Carbon to Nitrogen ratios increased with eye width, consistent with Weiser and Kaspari’s 337 (2006) finding that predatory genera have smaller eyes. The eyes of C:N predators were also 338 positioned closer to the side of their heads, suggesting that acute forward vision was not 339 critical for most predatory species in the study system (with Myrmecia as an obvious 340 exception). Contradicting this finding, δ15N predators (those with higher δ15N) had eyes 341 positioned higher on their heads. It is unclear why these two stoichiometric measures gave 342 opposing results, but the relationship between C:N and eye position (R2 = 0.36, p = 0.0004) 343 was much stronger than the marginally significant relationship between δ15N and eye position 344 (R2 = 0.14, p = 0.038) and was consistent with predictions. 345 Predatory ant species had greater physical protection than predominantly omnivorous species, 346 being both more sculptured (negative correlation with C:N) and more pilose (positive 347 correlation with δ15N). Cross-species variation in sculpturing and pilosity in arthropods is 348 poorly understood, but it is likely that these features provide protection from harsh 349 microclimates and/or attack by other animals. Carbon to nitrogen ratios might be related to 350 the ratio of protein (high nitrogen) and chitin (low nitrogen) in the exoskeleton or the cuticle. 351 Greater ‘sculpturing’ is associated with more complex, thicker cuticles, which might be more 352 protein-rich, and may provide better physical protection for predatory species. Pilosity may 353 have a similar protective function, with hairs reported to increase tolerance to dehydration 354 (Wittlinger et al. 2007). 355 Can we use morphological variation among species to predict preferred microhabitat? 356 Microhabitat variables measured at the level of the pitfall trap correlated with a range of 357 morphological traits when we accounted for phylogeny. Narrow heads, small mandibles, 358 dorsally positioned eyes and a lack of spines were associated with more open habitats. 359 Smaller heads and mandibles may be associated with more streamlined genera with faster 360 17 running speeds, such as Iridomyrmex, that dominate in open habitats (Gibb and Parr 2013) or 361 reduced dietary specialisation in simple habitats. Previous studies show that larger ants (Gibb 362 and Parr 2010; Arnan et al. 2013) and ants with relatively longer legs (Gibb and Parr 2010; 363 Wiescher et al. 2012) dominate in more open environments. However these relationships 364 were not significant at the microhabitat scale in this study. 365 Ant species with dorsal eyes used bare ground more than those with more laterally positioned 366 eyes, indicating that a broad visual field is likely to be more important in open habitats 367 (Weiser and Kaspari 2006; Gibb and Parr 2013), as may be a greater awareness of aerial 368 predators. Spinosity, which was highest in association with high canopy cover (and low bare 369 ground), might also be related to predation risk. However, little is known about how the 370 assemblage of predators of ants differs among habitats differing in complexity. 371 Do environmental variables at site scale predict trait sets for whole assemblages of ants? 372 The fourth corner analyses revealed a range of associations between morphological traits and 373 the environment at the level of assemblage. Assemblage-level patterns contrasted with the 374 patterns observed at the level of species. For example, none of the characteristics associated 375 with bare ground or canopy cover at the species level showed significant associations at the 376 assemblage level. Differences did not result from the use of phylogenetic independent 377 contrasts because relationships from standard regressions did not parallel those from the 378 fourth corner analysis. Differences in the scale at which species respond to their habitats may 379 have resulted in differences in trait-environment relationships between tests based on 380 microhabitat and macrohabitat. Alternatively, the failure of the species-based regressions to 381 account for the relative abundances of species and species and to therefore consider the 382 interactions leading to the observed relative abundances of species may limit our ability to 383 extrapolate from species- to assemblage-level traits. 384 18 At the site scale, ant species with larger body size (Weber’s length) were more prevalent 385 where there was more bare ground and less shrub cover. This contrasts with findings for 386 North American ants, for which ants in environments with less groundcover had relatively 387 longer legs but did not differ in size (Wiescher et al. 2012) but is in agreement with a range of 388 studies showing that body size or the body size index (the product of femur length and 389 pronotum width) is greater in open habitats (Sarty et al. 2006; Gibb and Parr 2010; Arnan et 390 al. 2013). Species with broader heads, including Myrmecia spp., Crematogaster spp., 391 Notoncus spp., Amblyopone australis and Tetramorium sp., were most common in sites with 392 little shrub cover and woody debris. 393 Polymorphic species were more prevalent in sites with bare ground and greater canopy cover. 394 This may reflect a high abundance of relatively large polymorphic Camponotine ants and 395 medium-sized polymorphic Melophorus in later successional habitats with open ground and 396 low shrub cover (Gibb and Cunningham 2013). Greater polymorphism in open (burned) 397 habitats was also detected by Arnan et al. (2013), who attributed the relationship to greater 398 variability in the response of individuals within a colony to temperature, which could increase 399 foraging time and therefore enhance colony success (Cerdá et al. 1997). A high prevalence of 400 pilose ants was also associated with bare ground, providing support to the hypothesis that 401 hairs increase tolerance to dehydration (Wittlinger et al. 2007). 402 Conclusions 403 We have shown that we can use morphology to make predictions about species function, 404 including trophic status and habitat use. These relationships were biologically plausible and 405 not driven primarily by distant phylogenetic relationships, despite strong differentiation of 406 subfamilies by morphology (Appendix II). This suggests that more recent evolutionary 407 pressures may drive many trait-environment relationships observed here. Further, it 408 19 reinforces that functional traits provide information on likely species biology over and above 409 what can be predicted by phylogeny alone. We also extended our test to the level of the 410 assemblage to examine how the environment regulates the prevalence of traits in an 411 ecological community and found that the prevalence of many traits reflects the environment. 412 We thus show that a trait-based approach to understanding community assembly is 413 achievable when integrating stoichiometry, morphology and phylogeny and will help to build 414 a more predictive framework for invertebrate ecology. However, we caution that a greater 415 understanding of the scale-dependency of these relationships is needed. 416 Ultimately, the functional trait approach should be applicable to assemblages across 417 biogeographic regions and even across multiple taxonomic groups, independent of the 418 phylogenetic history of the biota. To establish the generality of the trait associations we 419 measured here requires comparative studies considering phylogenetically distinct 420 assemblages (e.g., Gibb and Parr 2013). Experimental tests of specific traits across species 421 would also aid in verifying function. More broadly, another important early step is to extend 422 this technique from ants to include other epigaeic invertebrates, which are of similar size and 423 dwell in similar microhabitats and therefore face similar challenges (e.g., beetles, Barton et 424 al. 2011; and spiders, Langlands et al. 2011). New modelling techniques such as the fourth 425 corner analysis used here allow a more predictive approach, where relative abundances and 426 covariances of species can be better accounted for, although it is so far unable to account for 427 phylogeny. A predictive approach is particularly important in anticipating the biotic response 428 to the broad impacts of anthropogenic drivers of global change, such as habitat disturbance 429 and climate change (Andrew et al. 2013), both of which have significant impacts on the 430 ground-layer and thereby influence assemblages of epigaeic species. 431 20 Acknowledgements 432 We thank N. Banks, K. Pullen, E. Finlay and L. Garrett for help with field work and M. 433 Cosgrove for assistance in the laboratory. We are grateful to the farmers of Murrumbateman, 434 Bungendore and Boorowa, who allowed us to work on their land and Greening Australia for 435 providing contacts. S. Shattuck kindly provided ant identifications. This project was 436 supported through an OCE postdoctoral fellowship to HG and an ARC Discovery Project 437 (DP0985886) to DIW, NRA and HG. 438 Data accessibility 439 Data on which this paper is based will be archived in Figshare (www.figshare.com) following 440 acceptance for publication. 441 References 442 Ackerly DD, Cornwell WK (2007) A trait-based approach to community assembly: 443 Partitioning of species trait values into within- and among-community components. 444 Ecol Lett 10:135-145 445 Andrew NR et al. (2013) Assessing insect responses to climate change: What are we testing 446 for? Where should we be heading? PeerJ 1:e11 447 Arnan X, Cerda X, Rodrigo A, Retana J (2013) Response of ant functional composition to 448 fire. Ecography 36:1182-1192. doi: DOI 10.1111/j.1600-0587.2013.00155.x 449 Barton PS, Gibb H, Manning AD, Lindenmayer DB, Cunningham SA (2011) Morphological 450 traits as predictors of diet and microhabitat use in a diverse beetle assemblage. Biol J 451 Linn Soc 102:301-310. doi: DOI 10.1111/j.1095-8312.2010.01580.x 452 Bihn JH, Gebauer G, Brandl R (2010) Loss of functional diversity of ant assemblages in 453 secondary tropical forests. Ecology 91:782-792 454 21 Blackburn TM, Gaston KJ (1998) Some methodological issues in macroecology. Am Nat 455 151:68-83. doi: Doi 10.1086/286103 456 Blüthgen N, Gebauer G, Fiedler K (2003) Disentangling a rainforest food web using stable 457 isotopes: dietary diversity in a species-rich ant community. Oecologia 137:426-435 458 Brown AM, Warton DI, Andrew NR, Binns M, Cassis G, Gibb H (2014) The fourth-corner 459 solution – using predictive models to understand how species traits interact with the 460 environment. Methods Ecol Evol 5:344-352 461 Cerdá X, Retana J, Cros S (1997) Thermal disruption of transitive hierarchies in 462 Mediterranean ant communities. J Anim Ecol 66:363-374 463 Chown SL (2012) Trait-based approaches to conservation physiology: forecasting 464 environmental change risks from the bottom up. Philos T R Soc B 367:1615-1627. 465 doi: DOI 10.1098/rstb.2011.0422 466 Chown SL, Gaston KJ (2010) Body size variation in insects: A macroecological perspective. 467 Biol Rev 85:139-169 468 Cornwell WK et al. (2008) Plant species traits are the predominant control on litter 469 decomposition rates within biomes worldwide. Ecol Lett 11:1065-1071 470 Davidson DW (2005) Ecological stoichiometry of ants in a New World rain forest. Oecologia 471 142:221-231 472 DeNiro MJ, Epstein S (1979) Influence of diet on the distribution of carbon isotopes in 473 animals. Geochimica et Cosmochimica Acta 42:495-506 474 DeNiro MJ, Epstein S (1981) Influence of diet on the distribution of nitrogen isotopes in 475 animals. Geochimica et Cosmochimica Acta 45:341-351 476 Diamond SE et al. (2012) A physiological trait-based approach to predicting the responses of 477 species to experimental climate warming. Ecology 93:2305-2312 478 22 Eisner T (1957) A comparative morphological study of the proventriculus of ants 479 (Hymenoptera: Formicidae). Bulletin of the Museum of Comparative Zoology 480 116:429-490 481 Entling W, Schmidt-Entling MH, Bacher S, Brandl R, Nentwig W (2010) Body size-climate 482 relationships of European spiders. J Biogeogr 37:477-485 483 Fagan WF et al. (2002) Nitrogen in insects: Implications for trophic complexity and species 484 diversification. Am Nat 160:784-802. doi: Doi 10.1086/343879 485 Feener DH, Lighton JRB, Bartholomew GA (1988) Curvilinear allometry, energetics and 486 foraging ecology: a comparison of leaf-cutting ants and army ants. Funct Ecol 2:509-487 520. doi: Doi 10.2307/2389394 488 Felsenstein J (1985) Phylogenies and the comparative method. Am Nat 125:1-15 489 Fowler HG, Forti LC, Brandao CRF, Delabie JHC, Vasconcelos HL (1991) Ecologia 490 nutricional de formigas. In: Panizzi AR, Parra JRP (eds) Ecologia nutricional de 491 insetos. Editora Manole, Sa˜o Paulo, Brazil, pp 131-223 492 Freckleton RP (2009) The seven deadly sins of comparative analysis. J Evolution Biol 493 22:1367-1375. doi: DOI 10.1111/j.1420-9101.2009.01757.x 494 Funk JL, Cleland EE, Suding KN, Zavaleta ES (2008) Restoration through reassembly: plant 495 traits and invasion resistance. Trends Ecol Evol 23:695-703 496 Garland T, Ives AR (2000) Using the past to predict the present: Confidence intervals for 497 regression equations in phylogenetic comparative methods. Am Nat 155:346-364 498 Garland T, Midford PE, Ives AR (1999) An introduction to phylogenetically based statistical 499 methods, with a new method for confidence intervals on ancestral values. Am Zool 500 39:374-388 501 23 Gibb H, Ball JP, Atlegrim O, Johansson T, Hjältén J, Danell K (2005) The effects of 502 management on coarse woody debris volume and quality in boreal forests in northern 503 Sweden. Scand J Forest Res 20:213-222 504 Gibb H, Cunningham SA (2011) Habitat contrasts reveal a shift in the trophic position of ant 505 assemblages. J Anim Ecol 80:119-127 506 Gibb H, Cunningham SA (2013) Restoration of trophic structure in an assemblage of 507 omnivores, considering a revegetation chronosequence. J Appl Ecol 50:449-458. doi: 508 Doi 10.1111/1365-2664.12054 509 Gibb H, Parr CL (2010) How does habitat complexity affect ant foraging success? A test of 510 functional responses on three continents. Oecologia 164:1061-1073 511 Gibb H, Parr CL (2013) Does structural complexity determine the morphology of 512 assemblages? An experimental test on three continents. Plos One 8:e64005. doi: 513 10.1371/journal.pone.0064005 514 Greenslade PJM (1973) Sampling ants with pitfall traps: digging-in effects. Insect Soc 515 20:343-353 516 Harvey PH, Pagel MD (1991) The Comparative Method in Evolutionary Biology. Oxford 517 University Press, Oxford, UK 518 Kaspari M (1993) Body-size and microclimate use in Neotropical granivorous ants. 519 Oecologia 96:500-507 520 Langlands PR, Brennan KEC, Framenau VW, Main BY (2011) Predicting the post-fire 521 responses of animal assemblages: testing a trait-based approach using spiders. J Anim 522 Ecol 80:558-568 523 Liang KY, Zeger SL (1986) Longitudinal data analysis using generalized linear models. 524 Biometrika 73:13-22 525 24 McGill BJ, Enquist BJ, Weiher E, Westoby M (2006) Rebuilding community ecology from 526 functional traits. Trends in Ecology and Evolution 21:178-185 527 Michaud JP, Grant AK (2003) Intraguild predation among ladybeetles and a green lacewing: 528 do the larval spines of Curinus coeruleus (Coleoptera : Coccinellidae) serve a 529 defensive function? B Entomol Res 93:499-505. doi: Doi 10.1079/Ber2003269 530 Moreau CS, Bell CD, Vila R, Archibald SB, Pierce NE (2006) Phylogeny of the ants: 531 Diversification in the age of angiosperms. Science 312:101-104 532 Oster GF, Wilson EO (1978) Caste and ecology in the social insects. Princeton University 533 Press, New Jersey 534 Paradis E, Claude J, Strimmer K (2004) APE: analyses of phylogenetics and evolution in R 535 language. Bioinformatics 20:289-290 536 Ponsard S, Arditi R (2000) What can stable isotopes (d15N and d13C) tell about the food 537 web of soil macro-invertebrates? Ecology 81:852-864 538 R Development Core Team (2013) R: A language and environment for statistical computing. 539 R Foundation for Statistical Computing, Vienna, Austria 540 Renner IW, Warton DI (2013) Equivalence of MAXENT and Poisson point process models 541 for species distribution modeling in ecology. Biometrics 69:274-281 542 Sarty M, Abbott KL, Lester PJ (2006) Habitat complexity facilitates coexistence in a tropical 543 ant community. Oecologia 149:465-473 544 Shattuck SO (1999) Australian Ants: Their Biology and Identification. CSIRO Publishing, 545 Collingwood, VIC, Australia 546 Shipley B, Vile D, Garnier E (2006) From plant traits to plant communities: A statistical 547 mechanistic approach to biodiversity. Science 314:812-814. doi: DOI 548 10.1126/science.1131344 549 25 Silva RR, Brandão CRF (2010) Morphological patterns and community organization in leaf-550 litter ant assemblages. Ecol Monogr 80:107-124 551 Stoklosa J, Gibb H, Warton DI (2014) Fast forward selection for generalized estimating 552 equations with a large number of predictor variables. Biometrics 70:110-120. doi: 553 10.1111/biom.12118 554 Tillberg CV, Mccarthy DP, Dolezal AG, Suarez AV (2006) Measuring the trophic ecology of 555 ants using stable isotopes. Insect Soc 53:65-69 556 Weber NA (1938) The biology of the fungus-growing ants. Part 4. Additional new forms. 557 Part 5. The Attini of Bolivia. Revista de Entomologia 9:154-206 558 Weiser MD, Kaspari M (2006) Ecological morphospace of new world ants. Ecol Entomol 559 31:131-142 560 Westoby M, Wright IJ (2006) Land-plant ecology on the basis of functional traits. Trends in 561 Ecology and Evolution 21:261-268 562 Wiescher PT, Pearce-Duvet JMC, Feener DH (2012) Assembling an ant community: species 563 functional traits reflect environmental filtering. Oecologia 169:1063-1074. doi: DOI 564 10.1007/s00442-012-2262-7 565 Wilson EO (1953) The origin and evolution of polymorphism in ants. The Quarterly Review 566 of Biology 28:136-156 567 Wittlinger M, Wolf H, Wehner R (2007) Hair plate mechanoreceptors associated with body 568 segments are not necessary for three-dimensional path integration in desert ants, 569 Cataglyphis fortis. J Exp Biol 210:375-382. doi: Doi 10.1242/Jeb.02674 570 Yates ML, Andrew NR, Binns M, Gibb H (2014) Morphological traits: evidence of 571 predictable responses to habitat characteristics across bio-regions. PeerJ 2:e271 572 573 574 26 Figure captions 575 Figure 1: Significant phylogeny-adjusted relationships between morphological traits and 576 measures of diet. Each point represents the mean value for a species, adjusted for phylogeny. 577 Figure 2: Significant phylogeny-adjusted relationships between morphological traits and 578 measures of microhabitat use. Each point represents the mean value for a species, adjusted for 579 phylogeny. 580 Figure 3: Standardised interaction coefficient estimates for interaction terms from the fourth 581 corner analysis testing the relationship between morphological traits and the environment, 582 accounting for species abundances. Coefficients shown in red (positive) or blue (negative) 583 were significant in the best model. All trait variables except Weber’s length, sculpturing, 584 pilosity, polymorphism and the number of spines were residuals of regressions with Weber’s 585 length. 586 Figure 4: Contour plots of linear predictor values for each trait-environment interaction 587 selected by SIC (score information criterion) in the fourth corner analysis. Darker (redder) 588 areas correspond to larger predicted abundance across the trait-environment values. Circles 589 represent the partial residuals, where larger circles correspond to larger observed abundance. 590 27 Table 1: Morphological traits, hypothesised functions, measures taken and summary of significant associations from this study. Trait Hypothesised function Measure Regressions 4th corner PIC adjusted Unadjusted Weber’s length: Indicative of worker body size (Weber 1938), which correlates with metabolic characteristics Distance from the anterodorsal margin of the pronotum to the posteroventral margin of the propodeum - C:N - C:N + bare ground - shrub cover + slope Femur lengtha: Indicative of foraging speed, which reflects the complexity of the habitat (Feener et al. 1988) Length of the femur of the hind leg Head widtha: Size of gaps through which worker can pass (Sarty et al. 2006); mandibular musculature (Kaspari 1993) Measured across the eyes - bare ground - C:N - shrub cover - CWD Mandible length a and widtha: length, width and gape of mandibles relate to diet (Fowler et al. 1991); clypeus width indicates sugar feeding (Eisner 1957) The straight line distance from the insertion to the tip of the mandible - bare ground - C:N + δ15N Eye widtha: Eye size is indicative of food searching behaviour and activity times (Weiser and Kaspari 2006) Measured across the maximum width of the eye + C:N + C:N Eye positiona: Related to hunting method (Fowler et al. 1991) or the component of the habitat occupied (Gibb and Parr 2013) Head width across the eyes minus head width between the eyes – higher eye position indicates more dorsal eyes - canopy + C:N + δ15N +C:N Scape lengtha: Sensory abilities: longer scapes facilitate following of pheromone trails (Weiser and Kaspari 2006) Length of the antennal scape Sculpturing: Thickened, structured cuticles may increase dehydration tolerance 0 = no markings, shiny; 1= fine network of marks; cell-like shallow ridges; 2 = deeper dimples and ridging; 3 = surface heavily textured with ridges, grooves or pits - C:N - C:N - canopy Spinosity: Spines may act as an anti-predation mechanism (Michaud and Grant 2003) Count of spines on propodeum and petioles - bare ground + canopy Pilosity: Hairs may increase tolerance to dehydration or may relate to mechanoreception (Wittlinger et al. 2007) 0 = no or very few hairs; 1 = a sparse but regular covering of hairs; 2 = a consistent, moderate covering of hair; 3 = very dense hair covering + δ15N + bare ground Polymorphism: Different worker castes perform different tasks within the colony, allowing greater specialisation (Wilson 1953) 1 = monomorphic; 2 = polymorphic; 3 = dimorphic - δ15N + bare ground + canopy a As many size-related characteristics are correlated and we are interested in deviations from the expected value, these features were considered as residuals based on the regression with Weber’s length. 28 Table 2: R2 and significance for regressions with adjustment for phylogeny using phylogenetic independent contrasts and linear regressions testing the relationship between the morphological trait predictor variables and the diet-based response variables C:N and δ15N (n = 30) and microhabitat response variables percentage bare ground, canopy cover and leaf litter (n = 37). + 0.05