Forage collection, substrate preparation, and diet composition in fungus-growing ants



    Corresponding author
    1. Department of Biology, Centre for Social Evolution, University of Copenhagen, Copenhagen, Denmark
      Henrik H. De Fine Licht, Department of Biology, Centre for Social Evolution, University of Copenhagen, Universitetsparken 15, 2100 Copenhagen, Denmark. E-mail:
    Search for more papers by this author

    1. Department of Biology, Centre for Social Evolution, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author

Henrik H. De Fine Licht, Department of Biology, Centre for Social Evolution, University of Copenhagen, Universitetsparken 15, 2100 Copenhagen, Denmark. E-mail:


1. Variation and control of nutritional input is an important selective force in the evolution of mutualistic interactions and may significantly affect coevolutionary modifications in partner species.

2. The attine fungus-growing ants are a tribe of more than 230 described species (12 genera) that use a variety of different substrates to manure the symbiotic fungus they cultivate inside the nest. Common ‘wisdom’ is that the conspicuous leaf-cutting ants primarily use freshly cut plant material, whereas most of the other attine species use dry and partly degraded plant material such as leaf litter and caterpillar frass, but systematic comparative studies of actual resource acquisition across the attine ants have not been done.

3. Here we review 179 literature records of diet composition across the extant genera of fungus-growing ants. The records confirm the dependence of leaf-cutting ants on fresh vegetation but find that flowers, dry plant debris, seeds (husks), and insect frass are used by all genera, whereas other substrates such as nectar and insect carcasses are only used by some.

4. Diet composition was significantly correlated with ant substrate preparation behaviours before adding forage to the fungus garden, indicating that diet composition and farming practices have co-evolved. Neither diet nor preparation behaviours changed when a clade within the paleoattine genus Apterostigma shifted from rearing leucocoprinous fungi to cultivating pterulaceous fungi, but the evolutionary derived transition to yeast growing in the Cyphomyrmex rimosus group, which relies almost exclusively on nectar and insect frass, was associated with specific changes in diet composition.

5. The co-evolutionary transitions in diet composition across the genera of attine ants indicate that fungus-farming insect societies have the possibility to obtain more optimal fungal crops via artificial selection, analogous to documented practice in human subsistence farming.


The conspicuous ‘highways’ of Atta leaf-cutting ants carrying their forage to the nest are an impressive sight in neotropical habitats (Hölldobler & Wilson, 1990, 2008) and represent one of the most striking collective natural behaviours. Also the targets of this coordinated resource acquisition process are unusual, as the prime beneficiary is the fungus-garden symbiont of the colony, a highly specialised and co-adapted mutualist. However, it is mostly the scale of operations that may seem peculiar, because nutritional symbioses with microorganisms are in fact rather common, particularly among insects, because they allow the hosts to utilise nutritionally unbalanced food sources more efficiently (Klepzig et al., 2009). Almost all these mutualisms involve the unilateral or bilateral transfer of nutrients (Douglas, 1994; Stadler & Dixon, 2008) with gut symbionts and intracellular endosymbionts as classic examples (Suh et al., 2005; Moran, 2006). Fungus gardens, however, are an assembly of ectosymbionts dominated by a basidiomycete fungal clone that enhances the capability of its farming insect hosts to degrade structurally complex and nutritionally poor plant substrates (Paracer & Ahmadjian, 2000; Hölldobler & Wilson, 2008).

Behaviourally complex fungus agriculture has evolved independently nine times among the insects (Mueller et al., 2005): once in the fungus-growing termites (Macrotermitinae: Termitidae: Isoptera), which cultivate a single genus of basidiomycete Termitomyces fungi (Rouland-Lefevre, 2000; Aanen et al., 2002; De Fine Licht et al., 2005), once in the fungus-growing ants (Attini: Formicidae), which cultivate agaricaceous fungi (Mueller et al., 1998; Mueller et al., 2001), and seven times among the mycophagous bark beetles (Scolytinae: Curculionidae: Coleoptera). These beetle symbioses all arose among the ambrosia beetles, which cultivate a range of mostly Ophiostoma and Entomocorticium fungi (Harrington, 2005). A notable difference between these systems is that the fungus-growing ants and termites actively collect and incorporate new substrate in their tightly managed fungus gardens, whereas the bark beetle fungi grow in log galleries and degrade the surrounding wood under extensive agricultural management by the farming hosts, but without any direct substrate provisioning (Mueller et al., 2005). This implies that all fungus-farming insects have adaptations for proper habitat selection during colony founding, but that the fungus-growing ants and termites can also be expected to have adaptations for optimal forage acquisition and substrate processing that fungus-growing beetles lack (Harrington, 2005).

The origin, evolution, and natural history of the fungus-growing ant symbiosis have been extensively reviewed elsewhere (Mueller et al., 2001; Mueller, 2002; Hölldobler & Wilson, 2008; Mehdiabadi & Schultz, 2009). The symbiosis is generally divided into five major agricultural groups based on fungus-farming practices (Schultz & Brady, 2008): lower-attine agriculture, Apterostigma coral fungus agriculture, Cyphomyrmex yeast agriculture, higher attine agriculture, and leaf-cutting ant agriculture. Most research so far has focused on the behavioural aspects of Atta and Acromyrmex leaf-cutting ant foraging (Caldera et al., 2009), because these ants are most conspicuous and are important pests of agriculture and forestry (Cherrett, 1986; Schultz et al., 2005). However, leaf-cutting ant agriculture is an evolutionary highly derived state (for a recent review see Rico-Gray and Oliveira (2007)) comprising only ca 45 of the more than 230 known species of fungus-growing ants (Wirth et al., 2003; Schultz & Brady, 2008). They also represent only 2 of the 12 currently recognised genera [a number that will likely increase to 14 in the near future (Klingenberg & Brandao, 2009)]. This figure excludes the workerless parasitic genus Pseudoatta (Schultz & Meier, 1995), which is now considered to be a derived Acromyrmex (Sumner et al., 2004). The non-leaf-cutting ants are very different from the leaf-cutting ants in colony size, social complexity, mating system, and in type of substrate collected to manure fungus gardens (Mueller et al., 1998; Villesen et al., 1999; Mueller et al., 2001; Mueller, 2002; Villesen et al., 2002; Hughes & Boomsma, 2007; Schultz & Brady, 2008). Data on these attine ants have remained scarce and anecdotal, so that a systematic assessment of their diet composition and preparation behaviours is particularly needed (Leal & Oliveira, 2000).

The extensive garden management practices of fungus-growing ants has led previous authors to speculate that modifications in the symbiotic fungus could be the result of ant-imposed ‘artificial’ selection for more nutritious, more disease resistant, or easier to cultivate symbionts (Schultz et al., 2005). This notion is inspired by the fact that foraging ants control the flow of substrate material to their fungus gardens so that procedures for optimising yield would be similar to what is found in human subsistence farming. Such artificial selection has, for example, modified seed size and reduced stem branching in staple crops such as maize, rice and wheat (Konishi et al., 2006; Gregory, 2009) and led to the diversification of domesticated cattle into hundreds of divergent cattle breeds (Beja-Pereira et al., 2006). Likewise, animal husbandry has also influenced the frequency of human lactose tolerance in populations where adult ability to digest milk gave significant fitness benefits (Tishkoff et al., 2007). Although the early steps of human domestications may not all have been based on conscious decision making, it is clear that these processes increasingly became actively managed and culturally selected. However, none of this applies to fungus farming ants, so that the analogue of artificial crop selection is merely natural selection for reducing costs and increasing benefits during coevolution with an obligate mutualist. The key issue therefore is whether selective forces exerted by the host ants have modified traits of their fungal crops to match a different set of fitness requirements than the symbionts would be able to target or control on their own.

Here we review the available literature on foraging behaviour and substrate acquisition across extant genera of attine ants and we relate changes in diet composition to the presently known evolutionary history of the ants and their cultivars. We focus our analyses on broad-scale comparisons between the five major agricultural groups recognised by Schultz and Brady (2008), and we evaluate correlations between natural diet composition and the behavioural repertoire of the ants when preparing forage material for being incorporated in the fungus garden observed under laboratory conditions. We discuss how forage quality and availability may have affected coevolutionary transitions in attine fungus-farming and the extent to which analogies with artificial selection may apply.

Materials and methods

This review is based on an extensive search of the available literature on the diet composition of fungus-growing ants. It appeared that systematic quantitative surveys of diet composition are available mostly for the genus Atta and much more rarely for any other species or genus. In fact, most of the records obtained from other genera are from the first half of the last century, when such data were merely observations on forage particles that were published together with other natural history descriptions. Although they provide valuable biological detail, these data are often not directly comparable in a quantitative manner. We have therefore chosen to present and evaluate the data by scoring the number of records for each substrate category as a fraction of the total number of references available for each ant genus. We further adopted a procedure used by Schmid-Hempel (1998) for removing biases due to unequal intensity of study (in his case the social insect parasite literature being dominated by studies on honeybees), by plotting the number of substrate categories against the number of literature records available, and using the standardised residuals for further analysis. All records were carefully checked for cross-referencing to other studies, so that each observation that went into the final dataset represents a single independent observation for a given genus. Since the literature records do not measure diet composition in relation to the availability of potential food items in the environment, it was not possible to properly estimate niche breadths or calculate electivity indices (Feinsinger et al., 1981; Lechowicz, 1982).

Observations of substrate preparation behaviours were collected and compiled for each genus in a similar fashion to the diet composition data (Table 1). The behaviours for which observations were available were divided into nine categories: (1) antennation—antennal inspection of the substrate; (2) licking—first treatment of the substrate outside the fungus garden (including rasping with the glossae) before allowing contact with the fungus garden; (3) cutting/maceration—fragmentation of material into minute pieces; (4) chewing—the active mixing of substrate particles with saliva into a pulp-like mass; (5) metapleural gland (MG) grooming—rubbing of tarsi and metatarsi against the metapleural gland opening and active transfer of secretion to the mouthparts to be mixed with substrate; (6) licking inside the fungus garden; (7) faecal fluid—deposition of a droplet of faecal secretions on the substrate; (8) insertion in the garden—the substrate material is positioned and held in place in the fungus garden; (9) inoculation with fungal hyphae. This sequence of behaviours largely corresponds to the accurate account of substrate preparation behaviours in Acromyrmex octospinosus by Quinlan and Cherrett (1977) and the behaviours described by Weber (1972). However it is unclear what species of leaf-cutting ants Weber was describing, so that the Weber (1972) reference was not included in the analysis.

Table 1.  Categories of observed substrate preparation behaviours across the attine ants.
GenusSpeciesColonies observedBefore fungus gardenInside fungus gardenReferences
AntennationLicking outside fungus gardenCutting/macerationChewingMG groomingLicking within gardenFaecal fluidInsertion in gardenPlacement of fungal hyphae
  1. *Murakami and Higashi (1997) observed regurgitation of crop liquid in combination with licking behaviour within the garden and assume these substances are plant nectar used to manure the fungus garden.

  2. Behaviours are divided into ‘before’ and ‘after’ new substrate particles come into contact with the fungus garden. A ‘–’ indicates that a particular behaviour has not been studied, so that it could either be present (yes) or absent (no). A ‘?’ indicates that colony number observed are unknown. Five genera lack any data, but their names have been included to highlight gaps in knowledge.

Attasexdens rubropilosa3YesYesYesYesYesDe Andrade et al. (2002)
Attasexdens?YesYesYesYesYesNoYesYesMangone and Currie (2007)
Attacolombica?YesYesYesYesYesNoYesYesMangone & Currie (2007)
Attacephalotes?YesYesYesYesYesNoYesYesMangone and Currie (2007)
Attasp.?YesYesYesYesYesYesYesYesWeber (1958)
Acromyrmexoctospinosus?YesYesYesYesYesYesYesQuinlan and Cherrett (1977)
Acromyrmexoctospinosus?YesYesYesYesYesNoYesYesMangone and Currie (2007)
Acromyrmexoctospinosus?YesYesYesYesYesYesYesFernandez-Marin et al. (2003)
Acromyrmexhispidus?YesYesYesYesYesNoYesYesMangone and Currie (2007)
Acromyrmexechinatior?YesYesYesYesYesNoYesYesMangone and Currie (2007)
AcromyrmexLaticeps?YesYesYesYesYesNoYesYesMangone and Currie (2007)
Acromyrmexsp.?YesYesYesYesYesYesYesYesWeber (1958)
Trachymyrmexzeteki5NoNoYesYesYesYesYesYesMangone and Currie (2007)
Trachymyrmexbugnioni1NoNoYesNoNoNoYesNoMangone and Currie (2007)
Trachymyrmexcornetzi1NoNoNoNoYesNoYesNoMangone and Currie (2007)
Cyphomyrmexlongiscapus2NoNoNoNoYesNoYesYesMangone and Currie (2007)
Cyphomyrmexcostatus2NoNoNoNoYesNoYesYesMangone and Currie (2007)
Cyphomyrmexrimosus (yeast-growing)?YesYesYesYesYesWeber (1958)
Cyphomyrmexrimosus (yeast-growing)5NoNoYesYesYesYesMurakami and Higashi (1997)*
Myrmicocryptaednaella3NoNoYesYesYesMurakami and Higashi (1997)
Apterostigmadentigerum3NoNoNoNoYesNoYesYesMangone and Currie (2007)
Apterostigmacf. pllosum2NoNoNoNoYesNoYesNoMangone and Currie (2007)

To test for correlation between diet composition and substrate preparation behaviours, we constructed two distance matrices containing seven groups each: the genera Atta, Acromyrmex, Trachymyrmex, Cyphomyrmex, Myrmicocrypta, and Apterostigma, in addition to the C. rimosus species-group that differs from the rest of the genus by rearing their fungi as yeasts (Table 1). The matrix-wise correlation between the two comparative data sets was subsequently estimated using a Mantel test with 9999 permutations. The substrate distance matrix was constructed by recording the Euclidean distances d between the xth and yth observation:


using the arcsin square root transformed percentages of Fig. 1 where xj and yj are the respective data for the jth variable for group x and y (SAS statistical package ver. 9.1 for Windows). The substrate preparation behaviours where coded as binary variables (1 = behaviour observed and 0 = behaviour not observed), so that the distance matrix could be calculated by recording the direct number of differences in which a comparison of the same behaviour yielded a value of 0 (0 vs. 0 and 1 vs. 1) or 1 (0 vs. 1 or 1 vs. 0) after Huff et al. (1993) as implemented in GenAlEx6 (Peakall & Smouse, 2006). In this way the presence/absence of certain substrate categories in the diet is compared to the presence/absence of specific substrate preparation behaviours.

Figure 1.

Twelve categories of substrates used in the 12 extant genera of attine fungus-growing ants, with the number of currently recognised species in parentheses, based on Bolton's catalogue of ants of the world (Bolton et al., 2005), (Agosti & Johnson, 2005), and Fowler et al. (1986) and grouped according to the most recent phylogeny of Schultz and Brady (2008). The sizes of circles are proportional to the frequency of substrate use as derived from records in the literature. Colours define agricultural groups: leaf-cutting ants (green), basal higher attines (blue), lower attines (including both Paleoattini and Neoattini) (light brown), yeast-growing Cyphomyrmex (grey), and Apterostigma where most species cultivate pterulaceous coral fungi that are only distantly related to the leucocoprinous fungi reared by all other attine ants (dark brown). The genera Daceton and Orectognathus were the closest relatives of the fungus-growing ant tribe attini in the most recent phylogenetic study (Schultz & Brady, 2008).


Eighty-three references with a total of 179 records of diet composition were found and included in the dataset (see Table S1). Because the data were insufficient to cover species, we grouped them per ant genus, but with an additional subdivision to separate the grass-cutting species of Atta and Acromyrmex from the species collecting mostly leaves from dicotyledonous plants. Also the yeast-growing species of the Cyphomyrmex rimosus complex were kept separate from the remaining Cyphomyrmex. However, the Apterostigma data had to be pooled, since 75% (9/12) of the records did not report the species that was studied, so that it was impossible to distinguish between the pilosum species group cultivating pterulaceous fungi (majority of Apterostigma species) and the remainder of the genus cultivating agaricaceous symbionts (Fig. 1). Remarkably, some substrates such as flowers, dry plant debris, seeds, and insect frass are used by almost all extant genera of fungus-growing ants (Fig. 1), although the fraction of these specific substrates in the total diet differs greatly. Our analysis confirmed the general idea that only the ‘higher’ attines actively cut fresh plant material (not considering Mycetophylax, but see Fig. 1), but the fraction of this type of forage varies. Fresh plant material does not make up a larger fraction than dry plant material or insect frass in the total diet of Sericomyrmex and Trachymyrmex, whereas it does in Acromyrmex and, particularly, in Atta. However, the differences between non grass-cutting Acromyrmex and Trachymyrmex and Sericomyrmex are remarkably small (Fig. 1).

The number of substrate categories recorded was significantly correlated with the number of literature records available (Fig. 2a; p = 0.0297, r2 = 0.31), indicating that publication bias is real. There was a significant difference in the residual number of substrate categories between the five genus groups of fungus-growing ants: the lower-attine ants (including Apterostigma), the Cyphomyrmex yeast growers, the basal higher attine ants, the leaf-cutting ants and the grass-cutting leaf-cutting ants (Kruskal–Wallis: χ2 = 9.9, d.f. = 4, p = 0.0429), but this relationship became non-significant after omitting the highly specialised (single substrate use) grass-cutting ants from the analysis (Kruskal–Wallis: χ2 = 6.6, d.f. = 3, p = 0.087). This indicates that our adjusted data can be considered as being statistically unbiased (Fig. 2b).

Figure 2.

Comparative data of substrate categories and measures taken to make the data amenable for statistical analysis. (a) The number of substrate categories observed as a function of study intensity (number of records) across 15 groups of fungus-growing ants (after identifying grass-cutting and yeast-growing as additional groups) plotted on log scales to normalise variances (see Table S1 for details). The regression is: log Y = 0.43 + 0.49 log X, r2 = 0.31, N = 15; F1,13 = 5.96, P = 0.0297. (b) Means (± SE when n > 2) of studentised residuals (standardised using a t distribution) from the regression in (a). (c) The number of substrate categories as a function of the number of species per genus/group across the 15 groups of fungus-growing ants, plotted on log scales to normalise variances. The regression is: log Y = 0.36 + 0.47 log X, r2 = 0.37, N = 15; F1,13 = 7.75, P = 0.0156.

Both ‘higher’ attines and leaf-cutting ants have been recorded to collect a minimum of seven substrate categories (Fig. 1; excluding the grass-cutting species), similar to the lower-attine genera Cyphomyrmex and Apterostigma, which utilise eight and seven categories of substrate, respectively. However, Cyphomyrmex, Apterostigma, and Trachymyrmex are among the most species rich and geographically widespread genera of attine ants (Wheeler, 1907; Weber, 1972). This may have inflated the total number of forage categories (Fig. 1) as there is a significant positive relationship between number of species per genus and forage categories (Fig. 2c, p = 0.0156, r2 = 0.37). This seems understandable in light of species poor genera such as Mycetophylax, in the new sense of Klingenberg and Brandao (2009), being largely confined to arid regions and especially common in tropical coastal sand-dunes with limited possibilities for encountering diverse substrates (Diehl-Fleig & Diehl, 2007; Klingenberg et al., 2007; Klingenberg & Brandao, 2009).

After adjusting for differences in sample size there was no significant correlation between the number of substrate categories collected and the number of substrate preparation behaviours associated with incorporating substrate into the fungus garden (F1,5 = 2.8, p = 0.1567), but after exclusion of the specialised yeast-growing Cyphomyrmex (which only collect three types of substrates, Fig. 1) the correlation between substrate categories (adjusted for sample size) and preparation behaviours became significant (F1,4 = 22.25, p = 0.0092, r2 = 0.85). This was confirmed when we analysed the original distance matrices of number of substrate categories collected and number of specific substrate preparation behaviours expressed (Mantel test: r = 0.519, p = 0.0190). This result reflects that there is significant covariation between the forage that the ants collect in nature and the way in which they process material in laboratory colonies, where the full range of substrate items was not available (Weber, 1958; Quinlan & Cherrett, 1977; Murakami & Higashi, 1997; De Andrade et al., 2002; Mangone & Currie, 2007). It is also clear that the behavioural differences among farming groups (lower attines, higher attines, leaf-cutting ants) are more pronounced before forage particles enter the fungus garden than afterwards (Table 1).


Most fungus-growing ants are opportunistic foragers as substrates brought back to the nest range from fresh green leaves to insect frass, dead insects, and there is even a record of vertebrate tissue from a dead rodent being collected by Atta texana (Hölldobler & Wilson, 1990; Killion, 1991). The most narrow choices of diet are probably seen in the specialist grass-cutting species in the genera Atta and Acromyrmex living in open grassland habitats and almost exclusively using monocotyledon plants as substrate (reviewed in Fowler et al., 1986), and by members of the yeast cultivating Cyphomyrmex rimosus species group, which primarily use regurgitated liquid collected from nectar and insect frass as substrate (Murakami & Higashi, 1997). However, the remaining fungus-growing ant genera use an average of more than six different substrate categories (Fig. 1) that are highly variable in nutritional value for fungal growth.

Forage quality requirements

The value of any food item is determined by its nutritional value and the ease whereby it can be utilised, because certain food items may be difficult to digest, hard to find, or protected by defensive measures (Begon et al., 1996; Chown & Nicolson, 2004). Nitrogen content can be a limiting factor for tropical herbivorous insects (Mattson, 1980; Coley & Barone, 1996; Chown & Nicolson, 2004) and may therefore also be a limiting factor for fungus-growing ants and their symbionts (Mundim et al., 2009). Freshly cut leaf fragments have high nutritional value, and leaf-cutting ant fungus gardens respond to this by expressing high concentrations of extracellular amylase and proteases shortly after coming into contact with new substrate (De Fine Licht et al., in press). In addition to this abundantly available category of forage that the leaf-cutting ants specialise on, less abundant substrates such as insect frass and seeds also have a relatively high protein and nitrogen content (Chapman, 1998), which may explain why almost all extant genera collect these relatively rare substrates whenever they can. The fact that shed (but relatively fresh) flowers and fruit fragments are also collected by all genera for which data are available may have a similar explanation, but it seems more difficult to explain the collection of insect carcasses in Cyphomyrmex, Mycetophylax, Mycocepurus, and Myrmicocrypta (Fig. 1). This behaviour is also reported for some species of Trachymyrmex (Martin et al., 1969; Garling, 1979) and Apterostigma (Martin et al., 1969), but these observations could not be confirmed in the original references cited by Martin and Garling (Weber, 1958, 1966) and thus remain questionable. It is unclear whether the symbiont fungus is able to penetrate and digest the insect cuticle as the fungal hyphae only appear to form a mycelial mat on the surface of these insect fragments (Weber, 1972), which are often piled up next to the fungus garden without any direct physical contact or visible fungal growth (U. Mueller, pers. comm.). It is therefore possible that insect carcasses do not directly function as substrate for fungal growth, but somehow provide either the fungus or the ants with rare essential nutrients. However, if this would be the case, other genera that do not collect insect carcasses must acquire these essential nutrients from other sources.

Constraints in forage availability

Dry and fresh plant material offer different challenges for enzymatic digestion by the ants and their fungus garden. Dry plant parts may be very resistant to enzymatic degradation and require many specific glycoside hydrolases for degradation, which is clear from the potent saprophytic enzymatic machinery found in most lower-attine fungus gardens (De Fine Licht et al., in press). Fresh plant parts collected by leaf-cutting ants, on the other hand, can often be more easily degraded but may be protected against herbivory by defensive secondary plant compounds (Howard & Wiemer, 1986; Raven et al., 1999). Many studies, primarily for the genus Atta, have tried to document the influence of leaf surface waxes (Stradling, 1978), leaf toughness (Nichols-Orians & Schultz, 1990), leaf epiphylls (Mueller & Wolf-Mueller, 1991), leaf endophytes (Van Bael et al., 2009), and secondary plant compounds (Howard, 1987, 1988, 1990; Howard et al., 1989; Nichols-Orians, 1992) on ant diet composition. However, of the latter only plant terpenoids have been found to consistently inhibit Atta cephalotes workers from cutting leaves (Hubbell et al., 1983, 1984).

Long-term studies of diet composition in attine ants have revealed that choice of substrate is not necessarily constrained by availability (Farji-Brener, 2001), but often dependent on the season (Rockwood, 1976; Leal & Oliveira, 2000; Seal & Tschinkel, 2008). Seasonal shifts between wet and dry seasons in the (sub) tropics and between summer and winter in more temperate regions influence plant primary production and may remove most of the ant resource base for several months. In general, leaf-cutting ants cut more fresh leaves in tropical rainy seasons (Howard, 1987; Wirth et al., 1997, 2003) and are more opportunistic in dry seasons when more than 50% of the collected material may consist of non-green plant material such as stipules, fruits, seeds, and flower parts (Rockwood, 1975, 1976; Wirth et al., 1997). In an extensive study of the diet composition in non-leaf-cutting attine ants in Brazilian Cerrado vegetation, species of the genera Cyphomyrmex, Mycetarotes, Mycocepurus, Myrmicocrypta, Sericomyrmex, and Trachymyrmex all collected mostly flowers and fruits in the wet season, when these forage categories were abundant (Leal & Oliveira, 2000). However, during the dry season many more substrates were utilised, primarily insect frass, and insect carcasses by Cyphomyrmex, Mycetarotes, Mycocepurus, and Myrmicocrypta and vegetative plant parts, such as recently fallen leaflets and dry plant debris by Sericomyrmex and Trachymyrmex (Leal & Oliveira, 2000). Most attine ants thus appear to exhibit considerable seasonal variation in diet composition, which could not be accurately accounted for in the analyses of this review because the data were not sufficiently precise (Fig. 1).

Transitions in diet composition and substrate preparation behaviours

Not surprisingly the single most distinctive shift in diet is seen in the higher attines and especially in the leaf-cutting genera Atta and Acromyrmex, which are the only genera to actively cut fresh material from higher plants and to have leaves, flower petals, and fruits as the dominant substrate for their fungus gardens (Fig. 1). This is in agreement with a study by Wirth et al. (1997), which estimated that two-thirds of the total material harvested by Atta colombica over a 12-month period in a Panamanian rainforest was made up of green leaves. However, also T. cornetzi and T. diversus (members of the Trachymyrmex septentrionalis sister group to the leaf-cutting ants) have been observed to cut fresh leaves (Schultz & Brady, 2008), confirming that the genera Trachymyrmex and Sericomyrmex may generally have spectra of diet composition that are intermediate between the lower attines and the leaf-cutting ants. The complete transition to leaf-cutting agriculture may thus only have required rather minor morphological modifications in ant body size and adjustment of social communication to a larger scale (Wilson, 1980; Helantera & Ratnieks, 2008; Kelber et al., 2009). Carefully controlled studies involving switches between Trachymyrmex fungus garden symbionts and leaf-cutting ants, similar to Seal and Tschinkel (2007), could elucidate whether Trachymyrmex fungus gardens in fact contain the enzymatic machinery of normal leaf-cutting ant fungus gardens, and thus the ability to deal with the defensive secondary plant compounds of freshly cut plant material. Such experimental manipulations, exchanging fungus gardens between ant species, appears to be possible in several species, especially during the founding phase of colonies (Mueller et al., 2004; Advani & Mueller, 2006; Seal & Tschinkel, 2007; Poulsen et al., 2009).

Our comparative study found no clear difference in substrate use associated with Apterostigma ants, which is somewhat surprising as most species in this genus cultivate completely unrelated coral fungi in the family Pterulaceae (Munkacsi et al., 2004). However, the yeast growing symbiosis appears to be associated with clear differences in substrate use (Fig. 1) and substrate preparation behaviours, relative to, for example Myrmicocrypta ednaella (Murakami & Higashi, 1997). In particular, the use of regurgitated nectar is unique among yeast-growing Cyphomyrmex ants, as other lower attines may also collect nectar but generally do not seem to regurgitate it or implement it in the fungus garden. However these missing observations may be an artefact of low study intensity as few other lower-attine species have been studied in such detail as C. rimosus and M. ednaella (Murakami & Higashi, 1997) (Table 1). Lower-attine ants in general do not climb the vegetation in search for substrates (Hölldobler & Wilson, 1990), but C. rimosus in Panama sometimes nests in hollow twigs hanging from the understorey (H. H. De Fine Licht, pers. obs.), which could potentially facilitate encounters with nectar-bearing flowers.

Fungus garden substrates differ in texture and size and therefore likely require different ant preparation behaviours to reduce the risk of fungus garden contamination by unwanted microorganisms (Weber, 1958; Quinlan & Cherrett, 1977; Currie & Stuart, 2001; Mangone & Currie, 2007). The significant correlation between diet composition and substrate preparation behaviours (Fig. 1, Table 1) is consistent with this view and appears to be mostly due to the more elaborate cleaning behaviours of fresh leaves in the leaf-cutting ants, which are not seen in the lower-attine genera (Table 1). The specific characteristics of live plant material that likely require more elaborate cleaning behaviours may be the presence of defensive secondary plant compounds (Howard, 1987, 1988, 1990; Nichols-Orians, 1992), leaf surface waxes (Stradling, 1978), and endophytes (Van Bael et al., 2009) that may be hazardous for fungus gardens. On the other hand, leaf litter generally has a higher diversity of saprophytic fungi and bacteria than fresh leaves (Carlile et al., 2001), but this apparently does not require elaborate cleaning measures by the lower-attine ants M. ednaella and C. rimosus in comparison to leaf-cutting ants (Murakami & Higashi, 1997, Table 1). It would be interesting to know whether fungus-growing ants particularly select leaf litter fragments and other dry plant materials that are less colonised by microorganisms, so that more of the original resources remain. Questions like this underline the need for basic quantitative data on foraging under field conditions. Also detailed behavioural observations on substrate handling are lacking for many genera even though most of them can be kept in laboratory colonies where such behaviours could be fairly easily studied.

Does foraging and substrate preparation select for favourable fungal crop traits?

Lower-attine fungus agriculture evolved ∼ 50 MYA, but during the first ca 30 million years selection for symbiotic fungal traits was hampered by the repeated replacement of domesticated symbionts strains by free-living fungal stock (Schultz & Brady, 2008), which likely have prevented most fungal specialisations to a symbiotic lifestyle (Mueller et al., 1998; Dentinger et al., 2009; Vo et al., 2009). This is consistent with the diverse substrates of most lower-attine genera maintaining selection for a wide degradation spectrum, so that the main difference between free-living and domesticated fungi probably lies in the quality and spatial distribution of available substrates (Mueller, 2002; Schultz et al., 2005). Cooke and Rayner (1984) generally distinguish between fungal substrates according to two criteria: (i) whether the substrate can be readily assimilated or is difficult to degrade and (ii) whether the substrate is ‘continuous’ or ‘discontinuous’ (i.e. spatially patchily distributed) in its supply. By providing the symbiotic fungus with substrate, the fungus-growing ants effectively release the symbiotic fungus from the latter constraint by providing the symbiotic fungus with a continuous version of normally discontinuous substrates. We therefore expect that if any domestication effects are to be found in lower-attine symbionts, these would be modifications in the rate of degradation of normally discontinuous substrates, either upwards because of the more predictable supply, or downwards because the ants eliminate competing microbes.

About 20 MYA, the higher attine ants started to incorporate more fresh plant material after a single symbiont lineage had evolved gongylidia and became unable to live outside the obligate symbiosis, a process that ultimately led to the evolution of the leaf-cutting ants with their almost exclusive diet of freshly cut green leaves ca 10 MYA (Schultz & Brady, 2008, Fig. 1). The fungal cultivars, both of the lower and the higher-attine ants, are assumed to rarely reproduce sexually and are normally clonally and vertically transferred between ant colonies (Mueller, 2002; Mikheyev et al., 2006). The predominance of vertical transmission implies that selection on beneficial symbiont mutations is mostly indirect (via the survival of their host colonies) and slow (because recombination cannot combine independently evolved favourable traits) (Schultz et al., 2005). This provides a marked contrast with the fungus-growing termites where horizontal symbiont transmission has remained the norm, so that superior symbionts are both preferentially propagated within colonies and leave more sexual spores in the habitat for new colonies to collect (Aanen et al., 2009). However, recent studies suggest that leaf-cutting ants also have windows of opportunity for horizontal symbiont transmission during colony founding (Poulsen et al., 2009) and that this may have contributed to a recent selective sweep in fungal symbiont in response to the specific farming practices by Atta and Acromyrmex leaf-cutting ants (Mikheyev et al., in press). This implies that fungus-farming insect societies do have the possibility to obtain more optimal fungal crops via natural selection, but that the process is slow relative to the learning-driven cultural exchange of crops practised by human subsistence farmers (Diamond, 1998).


The authors thank Morten Schiøtt and two anonymous reviewers for comments on an earlier version of this manuscript. We were supported by the Danish National Research Foundation.