The cuticular profile of both Camponotus rufifemur and Crematogaster modiglianii highly differed from other, non-parabiotic Camponotus and Crematogaster species 9272829; unpublished data]. While there were only few aliphatic compounds with a chain length of C20-C33, both species possessed hydrocarbons of very high chain lengths (C35 up to C49, Figure 1) as well as steroids, which have not previously been detected on insect cuticles. The aliphatic profile of Crematogaster modiglianii consisted of hydrocarbons between C33 and C40. Beside n-alkanes and methyl-branched alkanes, more than 68% of its aliphatic cuticular compounds were unsaturated (Figure 1a, Tables 1, 2). Extracts of the body surface and postpharyngeal glands contained the same aliphatic substances in similar quantitative composition.

The Camponotus rufifemur surface profile mainly contained compounds beyond C₃₈, beside traces of lighter components. The two morphological varieties exhibit almost completely different surface profiles. The only substances in common were trace n-alkanes between C27 and C30 and C37-9-ene (Table 1). The red variety exhibited a highly unusual cuticular profile, 98% of the hydrocarbon quantities being methyl-branched alkenes. The main compounds, 27-MeC39-14-ene and 27-MeC39-16-ene, accounted for 88.7% of the total hydrocarbons. The other different methyl-branched alkenes were similar in respect to the positions of the methyl group and the double bond (Table 3). Chain lengths ranged from C38 to C41, with trace compounds between C24 and C37 (Figure 1b, Tables 1,2).

The profile of the black Ca. rufifemur variety consisted of even larger molecules, with 92.8% of the surface compounds between C44 and C49 (Table 1). At least 80% of the compounds were unsaturated (Table 2). Methyl-branched alkenes were also present, albeit not as abundant as in the red Ca. rufifemur variety. Minor compounds included n-alkanes, methyl-branched alkanes and aldehydes (Table 2). In both Ca. rufifemur varieties, PPG and surface extracts contained the same aliphatic compounds in similar relative quantities.

Besides aliphatic compounds, the surface profile of both ant species contained up to 24 components with a basic steroid structure (as inferred from mass spectra and diagnostic ions). Their mass spectra indicate a close chemical interrelatedness of the compounds. Due to the high substance quantities necessary for NMR analysis, their spatial molecular structure has not yet been resolved but is under investigation. Crematogaster modiglianii possessed high amounts of steroids on the body surface (2.59 ± 0.58 μg/worker, n = 11 colonies, mean and SE) which by far exceeded the hydrocarbons (0.48 ± 0.05 μg/worker, n = 11 colonies, mean and SE). In contrast, postpharyngeal gland extracts only contained minor amounts of steroids but high quantities of hydrocarbons. High steroid amounts of the same quantitative composition were also found in the Dufour gland, in separate alitrunk and gaster cuticular extracts and, albeit in lower amounts, in head cuticular extracts. They also occurred in cuticular extractions of living ants with SPME fibres, thus confirming that their presence in hexane extracts was not an artefact of concomitantly extracted glands. Altogether, Cr. modiglianii extracts contained 24 different steroid components with an abundance higher than 0.1% in at least one colony (percent of total steroid abundance). Their retention indices ranged between 20.38 and 25.77. Six of the 24 steroids were found in all Cr. modiglianii colonies in similar relative compositions. An additional eleven steroids were abundant in certain colonies but absent in others. The remaining seven steroids were irregularly found and never occurred in relative abundances higher than 1% (percent of total steroid abundance). Camponotus rufifemur extracts (both varieties) contained up to eight different steroids, all of which also occurred in Cr. modiglianii. The absolute steroid quantities in Ca. rufifemur were lower than the hydrocarbon quantities (black variety: 0.66 ± 0.22 μg steroids/worker and 1.79 ± 0.29 μg hydrocarbons/worker, n = 3 colonies; red variety: 0.41 ± 0.14 μg steroids/worker and 9.71 ± 3.79 μg hydrocarbons/worker, n = 4 colonies, mean and SE given).

Six hydrocarbons were shared between both parabiotic species. The red Ca. rufifemur variety shared three hydrocarbons with Cr. modiglianii. These were the two methyl-branched alkenes, 27-MeC39-14-ene and 27-MeC39-16-ene, which are the main constituents of the red Ca. rufifemur surface profile, and its saturated derivative, 13-MeC39 (Table 1). All three are absent in the black Ca. rufifemur variety. Cr. modiglianii colonies living with the red Ca. rufifemur variety (henceforth, 'red' Cr. modiglianii) exhibited significantly more 27-MeC39-14-ene and 27-MeC39-16-ene than those associated with the black variety (henceforth, 'black' Cr. modiglianii) (Mann-Whitney W = 30, p = 0.0043; N₁ = 5, N₂ = 6 colonies, Figure 2). The quantities of 13-MeC39 were not compared since they could not be separated from other methyl-branched C39 alkanes in Cr. modiglianii (Table 1). Traces of three other hydrocarbons common in Cr. modiglianii were detected in the black Ca. rufifemur variety (C35:1, C35, C37-9-ene, Table 1). Albeit the associated Cr. modiglianii possessed slightly more C37-9-ene than those living with the red variety, no significant differences were found.

Eight of the steroids common in Cr. modiglianii were also frequently found in Ca. rufifemur (inclusion criterion: median abundance > 0% in 11 colonies of both species; Figure 1d). Their relative abundances varied between parabiotic nests but were significantly correlated among the two species within a nest (Mantel test: r = 0.49, p = 0.041, N = 11; Bray-Curtis distances: 0.13 ± 0.08 (Ca. rufifemur), 0.43 ± 0.31 (Cr. modiglianii); mean and s.d.). A second Mantel test considered only three steroids with very similar mass spectra, which were present in all extracts (retention indices: 21.92, 22.24, 24.47; marked with asterisks in Figure 1d). This test yielded a highly significant correlation of steroid abundance among the two species of each parabiotic nest (r = 0.620, p < 0.001, N = 11; Bray-Curtis distances: 0.06 ± 0.04 (Ca. rufifemur), 0.13 ± 0.07 (Cr. modiglianii)).

Chemical differences between the two Ca. rufifemur varieties accounted for much of the variance in interspecific confrontations. In general, Cr. modiglianii workers tolerated only allocolonial Ca. rufifemur workers of the variety they were associated with. The focal Crematogaster modiglianii colony, which lived together with the red Ca. rufifemur variety, showed high aggression towards dead workers of the black Ca. rufifemur variety but not towards those of the red one (Figure 3). The generalized linear model (GLM) for total aggression explained 65.6% of the total deviance and yielded a highly significant effect of the Camponotus variety (58.7% explained deviance, Table 4). The remaining deviance could in part be attributed to differences between Camponotus colonies (p = 0.04), whereas the difference between intracolonial and allocolonial Camponotus was not significant (p = 0.12, Table 4). The non-parabiotic Camponotus (Tanaemyrmex) arrogans was attacked to a similar degree as the black Ca. rufifemur variety (Figure 3). When the analysis focused on the proportion of strong aggression only, the results were similar, with slightly stronger effects. Cr. modiglianii very rarely climbed onto the Ca. rufifemur bodies in this experimental series ('mounting behaviour').

In the arena confrontations, Cr. modiglianii was significantly more aggressive towards Ca. rufifemur from the respective other variety. The parameter 'within/across variety' explained 18.2% of the total deviance, followed by 'variety combination' (13.0% explained deviance), while 'intra-/allocolonial' did not explain a significant part of the deviance (Table 5). Cr. modiglianii workers frequently climbed on Ca. rufifemur bodies and walked around on them for up to one minute. This 'mounting behaviour' represented on average 18.6% of all interactions (Figure 4). The workers (especially in one of the two colonies) mounted Ca. rufifemur workers of their 'own' variety in significantly higher proportions (GLM for both colonies: F_{df = 1} = 6.85, p = 0.011) but did not otherwise differentiate between intracolonial and allocolonial Ca. rufifemur workers (F_{df = 1} = 0.14, p = 0.71).

In order to examine whether the differentiation between the colour varieties occurred in colonies in situ as well, we re-analyzed previous behavioural experiments reported in 2. Allocolonial aggression of Cr. modiglianii towards Ca. rufifemur was highly variable in this dataset, and we confirmed a high impact of the two chemical varieties on allocolonial aggression. The variable 'within/across varieties' (colonies A and B: black variety, colony C: red variety) explained 60.1% of the total variance of the data and was a clearly more powerful predictor than the differentiation between intra- vs. allocolonial combination (0.03% deviance explained, Table 6). 'Red' Cr. modiglianii colonies only attacked black Ca. rufifemur intruders and vice versa (Figure 5a). The highly significant impact of 'variety combination' (Table 6), however, showed that red Cr. modiglianii was more aggressive towards black Ca. rufifemur than black Cr. modiglianii towards red Ca. rufifemur.

In confrontations of Ca. rufifemur towards allocolonial Cr. modiglianii, Menzel et al. 2 had found low levels of aggression albeit they were higher than against intracolonial Cr. modiglianii. Similar to above, Ca. rufifemur workers were more aggressive towards Cr. modiglianii from the respective other variety (Figure 5b, Table 6).

To our knowledge, steroids have not been found in surface extracts of other ant species up to now, and to our knowledge have been found on insect cuticles only in one halictid bee 30. However, various Crematogaster species are known to have highly efficient poisons 3132. The genus Crematogaster has evolved a peculiar system of venom production which involves a cooperation of Dufour and poison gland. In several species the venom consists of precursors from the Dufour gland which are derivatized by enzymes from the poison gland 3334. Crematogaster poisons – from Dufour and poison glands, but also from hypertrophied metapleural glands – belong to such different chemical classes as cyclohexan derivatives, crematofuranes (cembranoid diterpenes), coumarin derivatives, alkylphenols, alkylresorcinols, salicylic acids, resorcylic acids, and polyfunctionalized long-chain derivatives 3335363738. Since extracts of Cr. modiglianii Dufour glands contained the same steroid composition as the body surface (but no other compounds), they are probably produced in this gland and then distributed onto the body surface. In Cr. modiglianii, steroid synthesis did not depend on biosynthetic precursors acquired from food. In two colonies kept in the laboratory for 15 and 6 months, respectively, the steroid profile did not change despite of an artificial diet of cockroaches, honey solution and Bhatkar diet (F.M. pers. obs.). Moreover, in one forest colony, the steroid profile remained relatively constant over three years, corroborating that the steroid composition is rather genetically determined than dependent on environmental factors.

It is notable that 98% of the entire hydrocarbon profile of the red Ca. rufifemur (and ≥ 13% of the black Ca. rufifemur hydrocarbon profile) were methyl-branched alkenes. This substance class seems to be generally very rare in insects and has been detected only in several Diptera and one Noctuid moth as pheromones 394041. Among ants, they have been found in traces in the ponerine ant Pachycondyla villosa and in two Leptothorax species 4243, but in higher abundances only in Nothomyrmecia macrops surface profiles, which is probably the most primitive existent ant species 44. That they make up almost the entire hydrocarbon profile is therefore highly unusual. Another unusual feature in both parabiotic species is the high hydrocarbon chain lengths. Although common in this study (Table 1), hydrocarbons beyond C₃₇ have not been found in non-parabiotic Camponotus and Crematogaster species 92728; unpublished data. Other studies report small concentrations of heavier hydrocarbons in other ant genera, but always accompanied by high amounts of lighter ones 4546. It is possible that extremely long-chain hydrocarbons are difficult to perceive by receptors and thus promote interspecific tolerance 2347. In one case, we observed that a non-parabiotic Cr. modiglianii colony was initially very aggressive against (black) Ca. rufifemur workers but treated them amicably (and had hence become habituated) after less than 24 h of exposure. Unsaturation in these long-chain hydrocarbons might be necessary to maintain a minimum fluidity of the cuticular profile 47.

Given the high allocolonial tolerance between parabiotic partners, the hydrocarbon overlap of the two species is surprisingly small. While the red Ca. rufifemur variety shared two compounds with its partner, the black variety only shared three trace compounds with Cr. modiglianii but otherwise possessed a completely different hydrocarbon profile. We tentatively suppose that Cr. modiglianii acquires 27-MeC39-14-ene 27-MeC39-16-ene from its red Ca. rufifemur partner although Ca. rufifemur generally tolerates Cr. modiglianii workers, including those lacking these substances 2. In a Cr. modiglianii colony kept in the laboratory without its previous red Camponotus partner, the compound disappeared from the profile after eight months of separation (F.M. pers. obs.). It is possible that the other hydrocarbons of the red Ca. rufifemur are acquired by Cr. modiglianii as well but remain beyond detectability due to their low abundances. The hydrocarbons of the black Ca. rufifemur, in contrast, were never found on Cr. modiglianii surface extracts. This is probably due to their high chain lengths, which makes the cuticular profile more solid and do not allow chemical transfer 47. In the light of the low overall hydrocarbon overlap among the two parabiotic ant species, chemical camouflage, a mechanism often found in social parasites 131415, must be dismissed as an explanation for mutual tolerance. However, the existence of only few substances common to both species might be a sufficient signal for tolerating the partner 48.

The steroid components, in contrast, showed high congruence among both species. We found that the relative composition of eight steroid compounds differs between colonies but is very similar among the two species of a parabiotic nest. Since it is highly improbable that Ca. rufifemur is able to synthetically copy the steroid profile of each respective partner colony, this result suggests that Ca. rufifemur acquires steroids from Cr. modiglianii. Notably, only a certain set of steroids is transferred to Camponotus, while others, despite of high abundance in Cr. modiglianii, were almost or completely absent from the Ca. rufifemur profile.

Two mechanisms seem possible for the observed transfer of chemical cues, namely trophallaxis and direct physical contact. Via trophallaxis, individual ants exchange not only food but also the PPG content, i.e. hydrocarbons relevant for nestmate recognition 49. The PPG of Cr. modiglianii indeed contained steroids, albeit in much lower concentrations than on the body surface, thus making trophallaxis a possible pathway for chemical transfer. Interspecific trophallaxis has been observed between the two parabiotic species (F.M. and A. Endler, pers. obs.) and also shown via stained food only fed to Cr. modiglianii (F.M., pers. obs.).

Another possible transfer mechanism is direct physical contact. We frequently observed that Cr. modiglianii climbed on living or dead Ca. rufifemur individuals (workers and alates). The latter sometimes tried to shake them off but did not show aggression. Though almost never observed in the field, this 'mounting behaviour' could be easily induced in the laboratory by keeping the two species separate for one or two days. Mounting may therefore represent another possible mechanism for transfer of surface chemicals.

The red and the black variety of Camponotus rufifemur are chemically distinct and – apart from trace compounds – do not share any hydrocarbons. The two dominant surface components of the red variety (substance #52, Table 1) are present in Crematogaster modiglianii colonies associated with this Ca. rufifemur variety but almost completely absent from those living with the black variety. Their abundance thus allows separating 'red' from 'black' Cr. modiglianii albeit the remaining surface profile is similar. The existence of two chemical Ca. rufifemur varieties accounts for most of the aggression variance in allocolonial encounters between the two species. Cr. modiglianii usually tolerated living or dead Ca. rufifemur workers of the same variety as their parabiotic partner but fiercely attacked those of the respective other variety (Figures 3, 4, 5, Tables 4, 5, 6). An analogous pattern was found in Ca. rufifemur. Despite of generally low aggression levels, black Ca. rufifemur workers were significantly more aggressive towards 'red' Cr. modiglianii workers than towards allocolonial 'black' Cr. modiglianii (Figure 5b). However, we did not detect a corresponding difference in the red Ca. rufifemur.

While much of the interspecific aggression can be explained by chemical differences, however, the low interspecific aggression within chemical varieties is still surprising. Rather than recognizing heterospecific nestmates, the two species seemingly recognize only the chemical variety of their partner and do not discriminate within these varieties. Nestmate recognition rather depends on volatile substances than on substances only perceivable through antennal contact 50. Due to their low volatility 47, very long-chain hydrocarbons are less detectable than short-chain molecules. Thus, olfactory receptors may additionally absorb traces of lighter hydrocarbons, thereby blurring inter-colony profile differences and hampering inter-colony discrimination 23. The role of the steroids in the nestmate discrimination process is still unclear and under investigation.

The high interspecific tolerance strongly contrasts with the South American parabioses of Crematogaster limata and the ponerine ant Odontomachus mayi, where the ants never tolerated heterospecific workers from foreign parabioses 3. In these associations, very low chemical overlap was found (no substance data given), suggesting that both species habituated to each other's colony-specific profiles. The associated Chilean species Camponotus morosus and Solenopsis gayi also showed distinct hydrocarbon profiles 51. In contrast to non-associated colonies, however, associated C. morosus had acquired small amounts of the S. gayi hydrocarbons. In both of these species, only individuals from associated colonies were tolerant towards allocolonial allospecifics 51, indicating that the acquisition of allospecific hydrocarbons promoted mutual tolerance.

The studies were conducted at Danum Valley Conservation Area from September to November in the years 2004 and 2007. Danum Valley represents one of the major remaining patches of tropical lowland rainforest in Sabah (Malaysian Borneo). The site has a typical equatorial rainforest climate with a mean annual temperature of 26.9°C and a yearly rainfall of 2700 mm. We studied parabiotic associations of Camponotus (Myrmotarsus) rufifemur Emery 1900 and Crematogaster (Paracrema) modiglianii Emery 1900. Their nests are commonly found in hollow, living tree trunks in the rainforest. Extracts of one parabiotic nest from the Kuala Belalong Field Studies Center (Brunei) were analyzed in addition.

Camponotus rufifemur occurs in two sympatric morphological varieties that have not previously been described (although Emery 52 notes that specimen from Sarawak are darker in colour than those from Sumatra). While one variety (henceforth, 'red' variety) has a reddish alitrunk and light red-brown legs, the other one (henceforth, 'black' variety) possesses a black alitrunk and dark red-brown legs. The area between the frontal carinae of soldiers is dull in the red but shining in the black variety. Although the ratio head width/scape length (in frontal view) tends to be higher in the large soldier caste of the black variety than in that of the red variety, no significant morphometric differences were found. In the following, we will refer to the varieties as 'red' and 'black' Ca. rufifemur. In order to allow a differentiation, the respective associated Cr. modiglianii will be called 'red' and 'black' Cr. modiglianii although we did not find morphological distinctions within this species. Voucher specimen of Cr. modiglianii and both Ca. rufifemur varieties are deposited at the Department of Zoology III, University of Würzburg and at the Forest Resarch Center in Sepilok, Sabah (Malaysia).

Extracts were prepared from both body surface and postpharyngeal glands (PPGs). For body rinses, 10 to 90 ants were killed by freezing and immersed in hexane for ten minutes. Extracts from single individuals contained quantities too low for reliable substance identification. Eleven parabiotic nests were sampled with one to eight (mean: 3.5) replicates per colony and species (ten from Danum Valley, one from Kuala Belalong). PPG extracts were obtained from three to four freshly dissected PPGs per sample dissolved in hexane. Octadecane (n-C₁₈) was used as internal standard in most samples. Cuticular substances were additionally obtained from living Cr. modiglianii workers brought into the laboratory in Würzburg with solid-phase microextraction (SPME). A SPME fibre (Supelco) coated with a 100 μm polydimethylsiloxan film was rubbed on the ant for 3 min and then directly injected into a ThermoQuest Trace GC.

Substances were identified by coupled capillary gas chromatography-mass spectrometry (GC-MS) with a Hewlett Packard 6890 series gas chromatograph coupled to a HP 5973 Mass Selective Detector. The GC was equipped with a J&W Scientific DB-5 fused silica capillary column (30 m × 0.25 mm ID; df = 0.25 μm). Temperature was kept at 60°C for 2 min then increased by 60°C/min up to 200°C and subsequently by 4°C/min to 320°C, where it remained constant for 10 min. Helium was used as carrier gas with a constant flow of 1 ml/min. A split/splitless injector was installed at 250°C in the splitless mode for 30 s. The electron impact mass spectra (EI-MS) were recorded with an ionisation voltage of 70 eV, a source temperature of 230°C and an interface temperature of 325°C. For analysis of hydrocarbons beyond C41, we used a DB-1 HT column (30 m × 0.25 mm ID; df = 0,25 μm). Temperature was raised from 60°C by 5°C/min up to 350°C and then kept constant for 10 min. The interface had a temperature of 350°C. All other settings were as above. The software MSD ChemStation (Version A.03.00) for Windows was used for data acquisition. We restricted the analyses to substances with a retention time beyond that of C19 since compounds with shorter chain length are likely to be too volatile to be relevant for nestmate recognition 56. Substances present in less than 50% of the samples are given in Table 1 (marked with *) but were disregarded from further analysis.

For quantification of steroid-like compounds and aliphatics shorter than C₃₃, we used ion counts from the GC-MS data and analysed both substance classes separately. Heavier hydrocarbons (beyond C33) were quantified using a high-resolution ThermoQuest Trace GC-FID with H₂ as carrier gas in order to achieve a better separation of the substances. We used a nonpolar capillary column [DB1 (J&W Scientific, Folsom, CA), 20 m × 0.18 mm, 0.18 μm film thickness] and the first temperature program given above (split closed for 30 s for extracts and for 2 min when using SPME fibers). The split/splitless injector port was kept at 260°C and the flame ionization detector (FID) at 340°C. Peak areas were computed with Chrom-Card 1.19 (CE Instruments, Milan, Italy). Mean absolute substance quantities were estimated by comparing substance peak areas with that of the internal standard (acquired with GC-FID) and dividing by the number of extracted individuals.

Profile similarities between the two partner species were analyzed for eleven parabiotic nests (including one from Kuala Belalong Field Studies Center). The average proportions of the steroid components per colony and species were calculated. The distances between colonies were calculated for each species separately using Bray-Curtis index of similarity and then compared between species using a Mantel test (1000 permutations).

Alkanes, methyl-branched alkanes and alkenes were characterized using diagnostic ions and retention indices calculated using Kovats' method 53. Unsaturated methyl-branched hydrocarbons were hydrated under a H₂ atmosphere using Palladium on activated carbon as catalyst to determine the position of the methyl group. The position of the double bond in methyl-branched and n-alkenes was determined using DMDS derivatization following 54. For methyl-branched alkenes, DMDS derivatization was insufficient for substance characterization since the position of the double bond relative to the methyl group remained unresolved and left two possible structures. Therefore, we cleaved the molecules in two parts at the position of the double bond via ozonisation. We diluted the sample in approx. 3 ml hexane, applied a constant flow of O₃ (300 mg/h) for ten minutes from a glass pipette (EO3G Ozone Generator, Easelec Technology Inc.) and directly injected the sample into the GC-MS. Ozonisation succeeded for substance 52 but not for the substances 37, 43, 56, and 61 (surface compounds of the red Ca. rufifemur, Table 1). However, it is highly probable that all methyl-branched alkenes are produced via the same biosynthetic pathway. We therefore tentatively inferred the position of the double bond from the structure of substances 52 (Table 1) and possibilities left from the DMDS results, which had succeeded for all of the above substances. Double bond positions in alkenes with chain lengths higher than C41 as well as in dienes and trienes could not be determined due to their low abundance and/or their high chain length, which resulted in derivatives which could not be detected using GC-MS. Aldehydes were identified by comparing their mass spectra to a commercial library (Wiley 275) and therefore remain tentative.

For the substances of the black Ca. rufifemur profile beyond C44, retention indices were calculated based on the retention times of an n-alkane standard (C21 to C40), C47 and C49, and therefore remain preliminary. These substances were identified based on mass spectra and hydrated samples. Unsaturation was further confirmed via fractionation using a SiOH column treated with AgNO₃. However, their characterization remains preliminary since the DMDS derivatized substances could not be detected using GC-MS.

We studied the reaction of Crematogaster modiglianii towards dead Ca. rufifemur workers from different colonies in Borneo. The reverse situation (Ca. rufifemur towards Cr. modiglianii) was not studied in this paper since Ca. rufifemur shows little discrimination between different Cr. modiglianii workers 2. A Cr. modiglianii colony (R0) had been collected in the forest circa one week prior to the experiments and was kept together with its red Ca. rufifemur partner in its original nest (a small tree trunk) in an open plastic box. The dead ants were placed onto the nest trunk with forceps such that several ants could interact with it simultaneously. During three minutes, each observed interaction was classified as peaceful (antennating), weakly (open mandibles) or strongly aggressive (biting or locking mandibles). An additional behaviour classified as peaceful was 'mounting', where the smaller Crematogaster (body length approx. 2–3 mm) climbed onto the Camponotus body (body length 5–13 mm). Continued interactions were recorded again after 10 s (the same behavioural classification as used in 2).

The aggressiveness of two other Cr. modiglianii colonies was estimated in arena confrontations. The workers had been collected in the forest one day prior to the tests and were kept in a plastic box among nestmates (but separate from the partner species) over night. Five Cr. modiglianii individuals were placed into a fluon-covered plastic cylinder (Ø 7.5 cm, height 5 cm) on top of a paper sheet floor. After 1 min to calm down, a dead Camponotus specimen was introduced. For the following 100 s we recorded the behaviour of the ants as above. Each living or dead ant was used for one assay only. In all of the above assays, we performed ten replicates per treatment.

From each replicate we calculated the proportions of all aggressive versus all non-aggressive interactions. Both strong and total (including weak) aggression were analyzed using generalized linear models (GLM) with quasibinomial error distribution and logit link function. In order to determine whether confrontations within and across chemical varieties differ, we used the according explanatory variable 'within/across variety' with two factor levels (which collapsed to 'Camponotus variety' in the first dataset). The variable 'variety combination' (with the factor levels 'black→black', 'black→red', 'red→red', and 'red→black') was nested in the former one. Further explanatory variables were 'colony combination' (nested in 'variety combination'), which collapsed to 'Camponotus colony' in the first dataset, and 'intra-/allocolonial'. Due to their nested structure, no interactions between the variables were possible. The impact of each variable was determined by likelihood ratio tests (F tests). We also re-analyzed data from 2 in a similar way, where we included the number of workers present in the experimental arena as explanatory variables (see 2 for details on the experimental setup). Since the statistical results for total aggression and for strong aggression only were similar, only the former will be reported in the results section. All computations were performed in R Version 2.5.1 55.