Unveiling community patterns and trophic niches of tropical and temperate ants using an integrative framework of field data, stable isotopes and fatty acids

Baiting

Fieldwork in Brazil was carried out in Florianópolis (Desterro Conservation Unit, 27°31′38″S, 48°30′15″W, altitude ca. 250 m), in December 2015 and January 2016, under sampling permit SISBIO 51173-1 (ICMBio), and export permits 15BR019038/DF and 17BR025207/DF (IBAMA). The vegetation consists of a secondary Atlantic forest with at least 60 years of regeneration. High rainfall rate along the coast results in high productivity, ant species richness and a tropical aspect for the Atlantic forest even at higher latitudes such as in our work (Silva & Brandão, 2014). In Germany, it was carried out in Darmstadt (Prinzenberg, 49°50′14″N, 08°40′01″E, altitude ca. 250 m), in July 2015 (no permits needed there). The vegetation consists of patches of mixed forest, beech forest and orchards, which were all covered by the sample grids.

Sampling design followed similar protocols in both sites (Table 1). Seven bait types were offered as proxies for resources that are widely used by ants in general (Kaspari, 2000; Blüthgen & Feldhaar, 2010; Lanan, 2014; for a full description of baits, see Supplemental Document S1). Sample points were distributed in grids and separated by 10 m. In each sampling session, only a single bait was offered per point, and bait types were randomized among points. Baits were set up in transparent plastic boxes and retrieved after 90 min. This procedure was repeated in different days until all bait types had been offered at each point (twice in Brazil). The design was based on Houadria et al. (2015) and evaluates use of multiple resources, differing from a typical cafeteria experiment, which is designed to assess preferred resources (Krebs, 1999).

Pitfall sampling

We performed a concomitant pitfall assessment to verify whether bait records represented well the epigeic community (Table 1). One vial per sample point was previously buried to avoid the digging-in effect (Greenslade, 1973), and replaced after collection for the next round. Pitfall and bait sampling were not performed simultaneously at the same point. Vials were buried at ground level, had six cm diameter and 150 ml volume, and contained 40 ml propylene glycol 50%.

Fatty acid analysis

In Brazil, samples were obtained from baits and complemented by colony sampling in November 2017. We only used ants from melezitose, sucrose and seed baits, to avoid interference of bait lipids. In Germany, they were obtained by colony sampling between July and August 2017. Samples were frozen at −18 °C directly from the field. Total lipids were extracted from the ants whole body using one ml chloroform:methanol solution, 2:1 (v/v). The solution was applied to SiOH-columns and the neutral parcel (= mono-, di- and triglycerides) eluted with four ml chloroform. Samples were analyzed with gas chromatography–mass spectrometry, following the same procedures described in Rosumek et al. (2017). NLFA amounts were obtained comparing their proportions to an internal standard (C19:0 in methanol; ρ_i = 220 ng/μl). Ants were subsequently dried to obtain their lean dry weight (= without lipids).

Stable isotope analysis

Ants collected in baits and conserved in ethanol 70% were used to analyze δ¹⁵N and δ¹³C. C and N isotope abundances were measured in a dual element analysis mode with an elemental analyzer coupled to a continuous flow isotope ratio mass spectrometer as described in Bidartondo et al. (2004). Relative abundances were calculated following the equation: δ_x = (R_sample/R_standard − 1) × 1,000 [‰], where R denotes the ratio between heavy and light isotopes of samples and international standards (N₂ in the air and CO₂ in PeeDee belemnite). Gasters were removed prior to analysis to avoid interference of gut content (Blüthgen, Gebauer & Fiedler, 2003).

Taxonomic identification

Ants were identified with taxonomic revisions, and comparison to identified specimens in collections and AntWeb images (AntWeb, 2016). Updated names were checked with Antcat (Bolton, 2018). Identifications were partially confirmed by taxonomists (see Acknowledgements).

In Brazil, ants were identified to genus level with Baccaro et al. (2015) and to species level with: Acanthognathus—Galvis & Fernández (2009); Acromyrmex—Gonçalves (1961); Cephalotes—De Andrade & Baroni Urbani (1999); Crematogaster—Longino (2003); Cyphomyrmex—Kempf (1965) and Snealling & Longino (1992); Gnamptogenys—Lattke (1995); Hylomyrma—Kempf (1973); Linepithema—Wild (2007); Octostruma—Longino (2013); Odontomachus and Pachycondyla—Fernández (2008); Pheidole—Wilson (2003); Wasmannia—Longino & Fernández (2007). Camponotus and Strumigenys were identified solely by comparison with collections.

In Germany, ants were identified to genus and species with Seifert (2007), Seifert & Schultz (2009) and Radchenko & Elmes (2010).

All material is stored in the collections of the Ecological Networks research group, Technische Universität Darmstadt, Darmstadt, Germany and Department of Ecology and Zoology, Federal University of Santa Catarina, Florianópolis, Brazil.

Data analysis

As a first step, we compared species’ incidences in baits and pitfalls (i.e., number of sampling points where it was recorded with each method). We assumed incidence in pitfalls to represent abundance in the community, and qualitatively compared it to incidence in baits to check whether common species were underrepresented in baits. To account for different efficiencies between methods, expected incidences were indicated by a line of slope m = I_baits/I_pitfalls, where I is the sum of all incidences for each method (Houadria et al., 2015).

Number of species, replicates and individuals per sample differed between methods, based on sample availability and ant size. In Brazil, we analyzed 24 species for baits, 41 for FAA and 31 for SIA. Method comparisons were performed only with 22 species considered in all three datasets. In Germany, seven species were analyzed with all methods. For a full list of recorded species and respective labels used in plots, see Table S1.

Unless noted otherwise, similarity matrices were based on unweighted Bray–Curtis dissimilarities, and Mantel tests and correlations used Spearman’s coefficient (rho). Analyses were run in R 3.4.3 (R Core Team, 2017) and PAST 3.14 (Hammer, Harper & Ryan, 2001).

For all bait analyses, we used proportion of occurrence on each bait type, relative to total records for each species. Only species with at least 10 records from five or more sample points were considered. In Brazil, day and night records were considered as independent to calculate proportions. For FAA, we calculated proportions of each NLFA relative to total composition, and used average proportions for each species. All NLFAs with average proportion >0.01% were considered. For SIA, we also used species’ averages and analyzed δ¹⁵N and δ¹³C separately, using Euclidean distances to build similarity matrices. A special case was Lasius fuliginosus in Germany, which was represented by a single colony that foraged over a large area. Bait records from different sample points were considered independent, and chemical results represent the average of different samples from that colony.

To analyze resource use, we used clustering and network analysis. UPGMA clustering was used for species, to show functional groups based on similar use of resources, and for baits, to show the structure of resource use in each community. Statistical significance of clusters was tested with SIMPROF (Clarke, Somerfield & Gorley, 2008) using the package “clustsig” (Whitaker & Christman, 2015). A Mantel test was used to compare similarity matrices of Brazil and Germany.

For network analysis, we used quantitative modularity (Q) (Dormann & Strauss, 2014) and specialization indices for species/resources (d′) and whole networks (H₂′) (Blüthgen, Menzel & Blüthgen, 2006), using the package “bipartite” (Dormann, Fruend & Gruber, 2017). Modularity shows how compartmentalized is a community, that is, if there are groups of species that strongly interact with groups of resources. In turn, d′ indicates whether individual species are specialized in certain resources, or resources that are used by a specialized group of species. H₂′ is an extension of d′ and shows how specialized the network is overall. H₂′ = 0 would mean that all species used resources in the same proportions, and H₂′ = 1 that each species has its exclusive pattern of resource use.

Specialization indices were also used to analyze species × fatty acids contingency tables. In this case, they indicate how exclusively NLFAs are distributed across species (Brückner & Heethoff, 2017). H₂′ = 0 would mean that all compounds occur in the same proportion in all species, and H₂′ = 1 that each species has its exclusive compounds. Correspondingly, relatively high d′ represents NLFAs that occur more exclusively in certain species, or species with more exclusive proportions of certain NLFAs. Low d′ means a compound that is widespread among species, or species with similarly generalized profiles. Additionally, we tested whether the two communities differed in their overall NLFA composition with PERMANOVA, using site as a fixed factor (Anderson, 2001). Homogeneity of multivariate dispersion was tested a priori with PERMDISP (Anderson & Walsh, 2013). To detect which NLFAs contributed to differences, we used SIMPER (Clarke, 1993). These tests were performed using package “vegan” (Oksanen et al., 2017).

To test whether niche breadths and NLFA profile diversity were different between communities at species level, we calculated Shannon diversity indices for each ant species as H′ = Σp_ilnp_i, where p_i is the proportion of each resource i used by the species, or NLFA found in its profile, and compared them with Mann–Whitney tests.

To test whether particular NLFAs were related to use of certain resources, we performed principal component analyses (PCA) using baits × species contingency tables, replacing zeros by small values (0.000001) and using centered log-ratio transformation to deal with the constant-sum constraint (Brückner & Heethoff, 2017). PC axes were correlated with NLFAs using function “envfit” from package “vegan.” We also compared proportions, amounts (in μg/mg; the amount of fat divided by lean dry weight) and unsaturation indices (UI; the sum of percentages of each unsaturated NLFA multiplied by its number of double bounds) between Brazil and Germany with Mann–Whitney tests. We did this for total fat and the three most abundant NLFAs (C16:0, C18:0 and C18:1n9). To test whether there was a direct relationship between total fat amount and C18:1n9, or total amount and UI, we correlated values for all individual samples of each community (166 in Brazil, 32 in Germany).

Finally, to test whether the results yielded by the three methods were correlated, we performed Mantel tests between similarity matrices of species for each method. We also correlated species’ d′ values for baits and NLFAs, to test whether specialization levels were related.

Use of resources

Most common species were recorded in baits in proportions similar to the expected, given their frequency in the community (Fig. S1 and Table S1). A few species were underrepresented in baits (e.g., Pachycondyla harpax, Myrmica scabrinodis, Stenamma debile), but, in general, species with few bait records were also rare in pitfalls. Thus, we consider that a representative part of the epigeic communities was properly sampled. Despite strong variation in total number of records, the number of species recorded in each bait was similar, with exception of large prey (Table 2).

Similarities in resource use were correlated between Brazil and Germany (Fig. 1, Mantel test, rho = 0.63, p = 0.03). In both communities, large prey was set apart from the other resources, being used less frequently and by fewer species. Seeds and melezitose changed positions between communities. In Germany, all ants used both sugars indiscriminately, while in Brazil several species used more sucrose (e.g., Camponotus zenon, Gnamptogenys striatula, Pachycondyla striata, Odontomachus chelifer, Solenopsis sp.6) and others used more melezitose (e.g., Pheidole aper, Solenopsis sp.8, Wasmannia affinis) (Table 2). Both modularity (Q_BR = 0.16, Q_GE = 0.14) and network specialization (H₂′_BR = 0.13, H₂′_GE = 0.12) were relatively low and similar between sites. Species used resources in different ways and a few were more specialized (see below), but there were no clear links between particular resources and species or groups of species.

In Brazil, W. auropunctata occupied a highly specialized niche, using only feces baits, which lead to the highest d′ values for any species and resource. Linepithema iniquum also showed a relatively higher specialization level due to its preference for dead arthropods and low use of sugars. P. striata and O. chelifer used more large prey, dead arthropods and sucrose. C. zenon grouped with them based on use of dead arthropods and sucrose, but avoided large prey. P. aper was the only species to have melezitose as its preferred resource. Other species showed higher redundancy and clustered together, including all Solenopsis and most Pheidole (Fig. 1; Table 2).

In Germany, only L. fuliginosus showed a relatively high specialization level and clustered separately, due to its almost exclusive use of animal resources (living prey and dead arthropods). Other species showed low specialization and dissimilarity (Fig. 1; Table 2).

Niche breadths were similar between communities (Mann–Whitney, p = 0.44). Average Shannon index was 1.6 ± 0.4 SD in Brazil and 1.7 ± 0.1 SD in Germany (Table 2).

Fatty acids

Temperate species contained much higher amounts of total fat than tropical ones (Fig. 2). Fatty acid compositions changed between communities (PERMANOVA, r² = 0.35, p < 0.01). Multivariate dispersion was heterogeneous, being higher in Brazil than in Germany (PERMDISP, F = 11.32, p < 0.01). This does not change the previous result because, in this case, PERMANOVA becomes overly conservative (Anderson & Walsh, 2013).

The main reason for this difference was the predominant role of C18:1n9 in temperate species (SIMPER, dissimilarity contribution = 47%, p < 0.01, Figs. 2 and 3; Table 3). In Brazil, composition was more balanced, which led to higher proportions of C18:0 (contribution = 21%, p < 0.01), although amounts were similar. C16:0 was proportionally the most abundant NLFA in Brazil and the difference from Germany was marginally significant (contribution = 20%, p = 0.06), although amounts again were higher in temperate species. A few other NLFAs had statistically significant, but very small contributions to the difference (Table S2).

Fatty acid compositions were generalized overall, but more homogeneous in Germany because of the predominance of C18:1n9 (H₂′_BR = 0.09, H₂′_GE = 0.03). Accordingly, NLFA profile diversity was higher in tropical species (average Shannon index = 1.5 ± 0.2 SD) than temperate ones (0.9 ± 0.2 SD) (Mann–Whitney, p < 0.01) (Fig. 3; Table 3).

In samples from Germany, there was no correlation between total amount of fat and percentage of C18:1n9 (rho = 0.19, p = 0.30) or unsaturation index (rho = 0.24, p = 0.89). In samples from Brazil, there was weak negative correlation between total fat and both C18:1n9 (rho = −0.16, p = 0.04) and unsaturation index (rho = −0.22, p > 0.01).

In Brazil, several fatty acids were related to resource use (Fig. 4; see Table S3 for PCA eigenvalues and full Envfit results). Species with higher C18:1n9 also used more dead arthropods (Envfit, r² of the NLFA with PC axes = 0.32, p = 0.02), while C18:2unk1 (an unidentified NLFA) was related to use of sucrose and large prey (r² = 0.31, p = 0.03). C14:0 (mystric acid) tended to be higher in species that used more seeds, feces and small prey (r² = 0.39, p = 0.01). Notice that the first two Principal Components explained only 60% of the variance and linear regressions were not strong. In Germany, most variation was along the sugar-protein axis. C18:0 and C17:0 (margaric acid) were strongly correlated with PC axes (r² = 0.84, p = 0.05 and r² = 0.78, p = 0.04, respectively). Both were higher in species that used more sugars, and C17:0 also was related to use of feces, although its relative abundance was very low in all species (<0.5%, Table 3).

Stable isotopes

In Brazil, W. auropunctata presented distinctive signatures for both isotopes. It was the species with highest δ¹⁵N, while most species ranged from 5.8 to 8.2, and six showed conspicuously lower signatures. Besides W. auropunctata, δ¹³C varied less, ranging from −24.1 to −27 (Fig. 5; Table 4).

In Germany, δ¹⁵N was lower overall, ranging from 3.6 (Lasius niger) to −1.1 (Lasius fuliginosus). Species varied little in δ¹³C (from −25.4 to −26.3), with values within the range of Brazilian species (Fig. 5; Table 4).

Isotope signatures were correlated for tropical species (rho = 0.43, p = 0.02). For temperate species, the correlation lacked statistical significance (rho = −0.46, p = 0.3), but their inclusion slightly strengthened the correlation for all species together (rho = 0.47, p < 0.01).

Dataset comparisons

In Brazil, similarities in bait use and NLFA profiles were correlated (Table 5). While δ¹⁵N similarities were correlated with similarities in bait use, but not with NLFAs, the opposite was found for δ¹³C. That is, similar use of resources among species was reflected in similar body fat composition, and both were related to their long-term trophic position, albeit in different ways. In Germany, no such correlations were found between datasets (although it was marginally significant for NLFAs and δ¹⁵N).

Exclusiveness (d′) of bait choices and NLFAs profiles were also correlated in Brazil (rho = 0.51, p = 0.02), suggesting that specialization in resources was reflected in more specific compositions of fatty acids. No correlation was observed in Germany (rho = −0.07, p = 0.86) (Fig. 6).

Use of resources

Resource use in Brazil was discussed in detail in Rosumek (2017), as well as the literature review on trophic niche of our identified tropical species. Large prey was the less used resource, because size and mobility of the prey limits which species are able to overcome them. Small prey and feces were also relatively less used, the first because also is relatively challenging to acquire, and the second probably due to smaller nutritional value. On the other hand, the other resources are nutritive and relatively easy to gather, particularly dead arthropods, which was by far the most used resource. Considering the similarity in resource use patterns, most general remarks in that work apply to Germany as well. The two main differences we found are discussed below.

The role of insect-synthesized oligosaccharides seems to be distinct between temperate and tropical communities. In Brazil, the aversion to melezitose showed by some species could represent a physiological constraint, since tolerance to oligosaccharides differs among ant species (Rosumek, 2017). For ants without physiological constraints, melezitose use might be opportunistic and does not necessarily mean that they interact with sap-sucking insects. However, honeydew is the only reliable source of oligosaccharides in nature, so the few species that preferred this sugar may engage in such interactions (particularly Pheidole aper). In Germany, on the other hand, all species used both sugars similarly. The two Myrmica, Lasius and Formica fusca are known to interact with sap-sucking insects, and Temnothorax nylanderi uses honeydew opportunistically when droplets fall on the ground (Seifert, 2007).

Seeds were other resource used differently, but this probably is consequence of our methodological choice of seeds with elaiosomes in Germany. Elaiosomes are thought to mimic animal prey and attract predators and scavengers (Hughes, Westoby & Jurado, 1994), not only granivores. Effectively, elaiosomes of Chelidonium majus are attractive to a wide range of ants (Reifenrath, Becker & Poethke, 2012). However, seeds were more extensively used in Brazil. A higher diversity of shape and sizes of seeds was offered there, which allowed more ants to use them.

Fatty acids

Fatty acid compositions were generalized, but differed between communities. In Germany, C18:1n9 plays a prominent role, making up for more than 70% of the NLFAs stored by ants. The amounts of fat also differed remarkably: in average, temperate ants stored over five times more fat. Similarly high amounts of total fat and percentages of C18:1n9 were observed in laboratory colonies of F. fusca and M. rubra (Rosumek et al., 2017), which suggest that it might be a general trend for temperate species. In Brazil, NLFA abundance at community level was more balanced between C16:0, C18:0 and C18:1n9. Both amounts and proportions of C18:1n9 were variable among species.

Organisms can actively change their fatty acid composition in response to environmental factors and physiological needs (Stanley-Samuelson et al., 1988). Temperature and balance between saturated and unsaturated NLFAs are important, because the fat body should present a certain fluidity that allows enzymes to access stored nutrients (Ruess & Chamberlain, 2010). C18:1n9 seems to be the only unsaturated fatty acid that ants are able to synthesize by themselves in large amounts (Rosumek et al., 2017). However, there was no positive correlation between amount of fat and C18:1n9 or unsaturation index of samples, which would be expected if C18:1n9 synthesis was a direct mechanism of individuals to balance saturation:unsaturation ratios (the weak negative correlation in Brazil also does not fit this hypothesis).

Therefore, we suggest that differences in C18:1n9 percentages and total amounts could be consequence of two distinct environmental factors. Under lower temperatures, higher proportion of unsaturated fatty acids is needed to maintain lipid fluidity (Jagdale & Gordon, 1997). Thus, temperate species might be adapted to synthesize and store more C18:1n9 to withstand the cold seasons. If this hypothesis were correct, the ants would maintain this high proportion throughout the year, since we collected in summer. In turn, high amount of fat could be a direct consequence of the marked seasonality in temperate regions. These species might be adapted to quickly acquire and accumulate energy reserves during the short warm season, while there is less pressure for this in regions where resources are available throughout the year.

We observed relationships between certain NLFAs and resources, although overall they were not strong and not necessarily result of direct trophic transfer. C18:1n9 was related to use of dead arthropods in Brazil. This NLFA is considered a “necromone,” a chemical clue for recognition of corpses by ants and other insects (Sun & Zhou, 2013), so it presumably increases in dead arthropods. However, only polyunsaturated fatty acids can be degraded to form C18:1n9 during decomposition, and C18:1n9 itself turns into C18:0 (Dent, Forbes & Stuart, 2004). Thus, for high C18:1n9 to be a direct result of scavenging, prey items should previously possess high levels of unsaturation. This might be an indirect effect as well: scavenger ants might be better at tracking and retrieving food items that are naturally rich in C18:1n9. No correlation was found in Germany, which could also be related to the special role of this NLFA in temperate species: its predominance due to environmental factors may override its dietary signal.

C18:2n6 occurs independently of diet only in very small amounts, and it is a potential biomarker (Rosumek et al., 2017). The differences we observed among species are direct result of diet. Its occurrence was more widespread in Brazil, but we observed no clear correlation with specific resources. C18:2n6 is found in elaiosomes, seeds and other arthropods in different amounts (Thompson, 1973; Hughes, Westoby & Jurado, 1994). Since it can come from different sources, C18:2n6 cannot be straightforwardly used as a biomarker for specific diets, but depends on a deeper analysis of the resources actually available in the habitat.

The biological significance of the correlations of NLFAs and resources in Germany is difficult to grasp. C18:0 does not appear to be preferably synthesized from carbohydrates, compared to C16:0 and C18:1n9 (Rosumek et al., 2017). Adding to the fact that such correlation was not found in Brazil, this might not represent a physiological link between sugar consumption and C18:0 synthesis. With low number of species in Germany, even strong correlations might be result of species-specific factors other than diet. The same might be said for C17:0, a fatty acid that occurs in very low amounts in several vegetable oils (Beare-Rogers, Dieffenbacher & Holm, 2001).

Interestingly, we did not observe any 18:3n3 or 18:3n6 (α- and γ-linolenic acids). Ants are not able to synthesize them, and they are assimilated through direct trophic transfer (Rosumek et al., 2017). In the studied communities, these fatty acids seem to be completely absent from food sources used by ants. This is an unexpected result, since their occurrence is well documented in elaiosomes and a wide range of insect groups that might serve as prey (Thompson, 1973; Hughes, Westoby & Jurado, 1994).

The use of fatty acids as biomarkers to track food sources is one of the greatest potentials of this method. However, it might be more suitable to detritivore systems, where the biomarkers are distinctive membrane phospholipids from microorganisms that decompose specific resources, and that end up stored in the fat reserves of the consumers (Ruess & Chamberlain, 2010). The NLFA profiles we observed are generalized, and the most relevant fatty acids could represent distinct sources and/or be synthesized de novo in large amounts. The biomarker approach might not be suitable at community level for ground ants, contrary to NLFA profiles (see Method Comparison below). However, it still might be useful to unveil species-specific interactions, or in contexts with less potential sources that can be better tracked (e.g., leaf-litter or subterranean species).

Stable isotopes

Trophic shift (i.e., the degree of change in isotopic ratios from one trophic level to another) varies among taxonomic groups and according to other physiological factors (McCutchan et al., 2003). “Typical” values of ca. 3‰ for δ¹⁵N and 1‰ for δ¹³C were experimentally observed in one ant species (Feldhaar, Gebauer & Blüthgen, 2010). Establishing discrete trophic levels is unrealistic in most food webs, particularly for omnivores such as ants (Polis & Strong, 1996), but species within the range of one trophic shift are more likely to use resources in a similar way. δ¹⁵N ranges of ca. 9‰ were observed for ant communities in other tropical forests, representing three trophic shifts (Davidson et al., 2003; Bihn, Gebauer & Brandl, 2010). This is similar to our range of 8.9‰ but, discounting W. auropunctata, the remaining range of 6.1‰ is more similar to what was observed in an Australian forest (7.1‰; Blüthgen, Gebauer & Fiedler, 2003). In Germany, only Lasius fuliginosus presented a distinct signature. In both communities, most species fell within the range of one trophic shift.

δ¹³C showed smaller, but meaningful, variations that were correlated to δ¹⁵N. δ¹³C is less applied to infer trophic levels, as it is more sensitive to sample preservation method and diet composition (Tillberg et al., 2006; Heethoff & Scheu, 2016). An average change of 0.61‰ was observed in samples stored in ethanol by Tillberg et al. (2006). However, we observed correlations (including with NLFAs—see below) despite this eventual change, and it would not affect the similarity among species and between communities. Primary consumers using distinct plant sources may present differences of up to 20‰, and this will influence the signature of secondary consumers (O’Leary, 1988; Gannes, Del Rio & Koch, 1998). However, in our case, only W. auropunctata presented such distinct value.

Again, both isotopes suggest that the core of these communities is composed by generalists that broadly use the same resources. Since we did not establish baselines, lower values in Germany do not necessarily mean lower trophic levels in this community. Isotope signatures for the same species are highly variable among sites in Europe (Fiedler et al., 2007), and this variation can be the result of either different isotope baselines or actual changes in species’ ecological roles.

Low δ¹⁵N suggest that a species obtain most of their nitrogen from basal trophic levels, mainly plant sources (Blüthgen, Gebauer & Fiedler, 2003; Davidson et al., 2003). This fits the six species with lowest δ¹⁵N in Brazil. Two were fungus-growing ants (Acromyrmex aspersus, Trachymyrmex sp.1), which use mostly plant material to grow its fungus. The others were species that forage frequently on vegetation, besides the ground (Camponotus lespesii, Camponotus zenon, Crematogaster nigropilosa, Linepithema iniquum). Arboreal species that heavily rely on nectar or honeydew usually present low δ¹⁵N, which may be the case for these species. Linepithema represents well this trend: the two mainly ground-nesting species, Linepithema micans and Linepithema pulex, presented higher signatures than the plant-nesting Linepithema iniquum (Wild, 2007).

Community patterns and method comparison

The correlations we observed between methods are interesting from both the methodological and the biological perspective. From a methodological viewpoint, for terrestrial animals, this is the first time an empirical relationship is shown between patterns of resource use and composition of stored fat in natural conditions, and that both relate to their long-term trophic position. Although differences between species were small, these relationships were robust enough to be detected by different methods. From a biological viewpoint, it highlights several physiological mechanisms involved in such relationships. We will discuss in the following some of these mechanisms, as well as caveats that are often cited for these methods. They probably still influence our results and correlations, but did not completely override the patterns.

A commonly cited caveat for using baits is that ants could be attracted to the most limited resources, instead of the ones they use more often. Evidence for this comes mainly from nitrogen-deprived arboreal ants (Kaspari & Yanoviak, 2001), and some cases are discussed below (see method complementarity). However, this effect might be less pronounced in epigeic species, and our results suggest that there is convergence between bait attendance, and medium- and long-term use of resources.

Diet may significantly change NLFA composition in a few weeks (Rosumek et al., 2017) and persist for a similar time (Haubert, Pollierer & Scheu, 2011). Therefore, the “snapshot” of resource use we observed with baits should represent at least the seasonal preferences of the species. A seasonal study on NLFA compositions can bring valuable information on resource use changes, or if they are stable throughout the year.

Adult ants are thought to feed mostly on liquid foods, due to the morphology of the proventriculus, which prevents solid particles to pass from the crop to the midgut (Eisner & Happ, 1962). Larvae are able to process solids and possess a more diversified suit of enzymes, and are sometimes called the “digestive caste” of the colony (Hölldobler & Wilson, 1990; Erthal, Peres Silva & Ian Samuels, 2007). Trophallaxis is an important mechanism of food sharing between workers and larvae. Our results suggest that the trophic signal of NLFAs is not lost in this processes, and that might be true even for solid items such as arthropods or seeds. However, the similarities could be as well the result of direct digestion and assimilation of liquid sources (sugars, hemolymph).

We also found correlations with stable isotopes. They were weaker than between baits and NLFAs, and different for each isotope. For δ¹⁵N, it shows that patterns of resource use are more correlated with trophic level. Protein amino acids must be obtained from diet or synthesized from other nitrogenated compounds, so the signal relative to nitrogen sources should be more preserved. This also fits to the idea that δ¹⁵N reflects larval diet, because it is in this stage that ants grow and build most of their biomass (Blüthgen & Feldhaar, 2010). On the other hand, it makes sense that the signal relative to carbon sources is related to NLFA composition. We should note that we removed the gaster of the ants used in SIA, so we observed only the signal of carbon incorporated in the other body parts. This is related to dietary carbon, but a stronger signal could be expected if the fat body is included.

The low source-specificity of stable isotope signatures might also lead to relatively weak correlations. Pachycondyla striata and O. chelifer had the same δ¹⁵N despite their different preferences. Other species that appear to be mostly scavengers had similar or higher δ¹⁵N than those two “predators,” such as Linepithema micans, Pheidole sarcina, Pheidole lucretii and Pheidole sp.4.

In Germany, no correlation was observed between methods. This is probably a consequence of the low number of species available in the community. The relationships found in Brazil might be valid for other communities, although ecological context and physiology might change their significance or strength.

Species niches and method complementarity

Niche differences were correlated at community level, but the use of different techniques allows better understanding of species’ niches. Method complementarity is particularly important if one is interested in the functional role of individual species, not only in overall patterns. Some cases are described below.

In Brazil, W. auropunctata was distinct from the remaining community, both in resource use and isotopic signature (unfortunately, no NLFA samples were obtained for this species). Strong preference for feces is a novel behavior for this species, known to invade and dominate disturbed habitats, but less dominant inside forests (Rosumek, 2017). Its isotopic signature confirms that they have a highly differentiated diet, and could be direct result of a feces-rich diet. In herbivorous mammals, feces are usually enriched in δ¹⁵N relative to diet (Sponheimer et al., 2003; Hwang, Millar & Longstaffe, 2007). The proposed mechanism of ¹⁵N enrichment along trophic levels states that this happens due to preferential excretion of ¹⁴N, and it is assumed that most nitrogen is excreted in the urine, which is depleted in ¹⁵N (Peterson & Fry, 1987; Gannes, Del Rio & Koch, 1998; but see Sponheimer et al., 2003). However, ¹⁵N-enriched feces were also observed in uricotelic organisms, such as birds and locusts (Webb, Hedges & Simpson, 1998; Bird et al., 2008). Thus, high δ¹⁵N is consistent with a diet based on ¹⁵N-enriched feces from other consumers. The relationship with the δ¹³C signature is less clear, but it also suggests high specialization.

This behavior might be a local adaptation, but also could indicate that W. auropunctata shifts to less disputed resources inside native forests (although the exact resources used may be context-dependent). In Davidson (2005), Wasmannia species (including W. auropunctata) presented relatively high δ¹⁵N and were considered highly carnivorous. However, our result shows that high δ¹⁵N should not be taken from granted to represent high-level consumers. It might be a solid generalization for communities, but other trophic pathways may lead to such signatures. Due to their lack of specificity, isotope signatures should be combined with field observations to provide reliable information at species level. As another example, the second highest δ¹⁵N in our work was observed in Pheidole sp.7, a species that used mainly seeds and was seldom recorded in animal (or feces) baits.

Another example where results seem to be contradictory is L. fuliginosus, which showed strong preference for animal baits, but low δ¹⁵N. In this case, the natural history of the species is well known, and it strongly interacts with aphids, particularly the giant oak aphid Stomaphis quercus (Seifert, 2007). This suggests that this aphid’s honeydew is not enriched in ¹⁵N and has a composition similar to the plants on which they feed. The honeydew supply should be abundant, since ants basically ignored sugar baits, but also relatively poor in nitrogen, which makes L. fuliginosus use animal sources whenever possible. A similar pattern may apply to Linepithema iniquum in Brazil, which also combined low δ¹⁵N with preference for animal baits. This species is also known to use extra-floral nectaries and honeydew (Rosumek, 2017), but does not have such strong and specific interactions as its temperate counterpart.