The evolution of mutualism


Egbert Giles Leigh, Jr., Smithsonian Tropical Research Institute, MRC 0580-02, Unit 9100, Box 0948, DPO AA 34002-9998, USA. Tel.: +507 212 8940; 507 212 8900; Fax: +507 212 8937; e-mail:


Like altruism, mutualism, cooperation between species, evolves only by enhancing all participants’ inclusive fitness. Mutualism evolves most readily between members of different kingdoms, which pool complementary abilities for mutual benefit: some of these mutualisms represent major evolutionary innovations. Mutualism cannot persist if cheating annihilates its benefits. In long-term mutualisms, symbioses, at least one party associates with the other nearly all its life. Usually, a larger host harbours smaller symbionts. Cheating is restrained by vertical transmission, as in Buchnera; partner fidelity, as among bull-thorn acacias and protective ants; test-based choice of symbionts, as bobtail squid choose bioluminescent bacteria; or sanctioning nonperforming symbionts, as legumes punish nonperforming nitrogen-fixing bacteria. Mutualisms involving brief exchanges, as among plants and seed-dispersers, however, persist despite abundant cheating. Both symbioses and brief-exchange mutualisms have transformed whole ecosystems. These mutualisms may be steps towards ecosystems which, like Adam Smith’s ideal economy, serve their members’ common good.


Competition brings forth mutualistic interdependence in human society (Smith, 1776). Likewise, the competitive process of natural selection drives the evolution of mutualism in ecological communities (Darwin, 1862). Indeed, mutualisms are more frequent where competition for food (Wilkinson, 1987) or pressure from predators (Degnan et al., 2009) or herbivores (Palmer et al., 2008) is more intense. What sets of factors favour the evolution of mutualism? How do they bring it about? How have mutualisms affected ecosystem evolution? Simple, general principles governing the evolution of mutualism have proved elusive (Herre et al., 1999), although, like rocks exposed by a falling tide, some preconditions for the evolution of mutualism are becoming progressively clearer (Frank, 1994; Sachs et al., 2004; Foster & Wenseleers, 2006; Sachs, 2006; Fletcher & Doebeli, 2009). How mutualisms affect ecosystems is a newer question (Leigh, 1999; Douglas, 2010).

This paper draws heavily on analogies with human economies. These analogies focus on the economies of Adam Smith’s (1776) day, where ecological analogies are more obvious, rather than today’s globalized behemoths. In nature, organisms compete for resources to live and reproduce. In human economies, people compete for resources to live as comfortably, and to settle their children as securely, as possible. In both, competition drives adaptive change. In nature, the differential spread of successful competitors’ genes drives this change; in human economies, the differential spread of knowledge of how to make more successful livings does so. In both, competition favours innovation, exploiting new resources or opportunities, or using old ones better. In nature, competition favours the diversification of species; among humans, the diversification of occupations—ways of making a living. To exploit new opportunities or compete better with third parties, people cooperate with other people of similar, or of different occupations, and sometimes with members of different species, as in agriculture or pastoralism. In nature, many organisms compete better by cooperating with members of the same, or other, species. Most cooperation and mutualism involves exchange of goods and services, or tokens enabling their procurement (Smith, 1776; Schwartz & Hoeksema, 1998). In nature and among humans, competition and diversification tend to increase the intensity of competition, pace of life, degree and variety of interdependence, and productivity (Leigh & Vermeij, 2002; Leigh et al., 2009). Despite the contrasting mechanisms of inheritance in nature vs. in human economies, and the far more intricate and wide-ranging cooperation enabled by human intelligence, speech and technology, the prevalence of competition leads to fundamentally similar trends in economies and ecosystems (Vermeij, 2004, 2009; Leigh et al., 2009).

There are two basic types of mutualism among species (Table 1). The type less discussed (because it is rarer?) involves members of different species joining in a single act or strategy that is more effective when more individuals employ it. Such mutualisms do not depend on division of labour among participants. Examples include gregarious fruiting of many tree species in dipterocarp forests, which reduces the proportion of seeds lost to seed-eaters (Sun et al., 2007), and Müllerian mimicry among distasteful butterflies that reduces the cost of educating predators about a warning coloration that advertises distastefulness (Fisher, 1930; Gilbert, 1983; Elias et al., 2008; Sherratt, 2008).

Table 1.   Classifying mutualisms.
I. By-product mutualisms
II. Mutualisms where each partner has behaviours selected to benefit the other
 A. Mutualisms without division of labour
  1. Mutualisms of mutual benefit with no possibility of cheating: each participant benefits itself and others by sharing in a common action [gregarious fruiting in dipterocarps to satiate seed predators: Sun et al., 2007)]
  2. Mutualisms whose participants share in a common action offering scope for cheating [Mullerian mimicry among butterflies to simplify education of predators, parasitized by palatable Batesian mimics: Fisher, 1930]
 B. Mutualisms with division of labour
  1. Long-term mutualisms (symbioses)
   a. Mutualisms enforced by transmission of symbionts from a host to its offspring [organelles in eukaryote cells: Eberhard, 1980]
   b. Mutualisms enforced by partner fidelity [sponges that fuse to pool capacities to resist different hazards for their common good: Wulff, 1997, 2008]
   c. Mutualisms enforced by partner choice [bobtail squid that test bioluminescent bacteria before admitting them as symbionts: Nyholm & McFall-Ngai, 2004]
   d. Mutualisms enforced by partner sanctions [legumes and their nitrogen-fixing bacteria: Kiers et al., 2003]
  2. Mutualisms of brief exchange
   a. Mutualisms enforced by partner sanctions [cleaner fish and their clients: Bshary & Grutter, 2005]
   b. Mutualisms with a limited degree of partner choice [seeds and their dispersers: Howe, 1986]

More often, mutualism between species involves the exchange of different goods or services, each reflecting the particular aptitudes of the species providing it, for their common good. A human counterpart is trade between people living by different occupations. Trade is advantageous if two groups need the same two products, but one makes the first better or cheaper than the second product, whereas the reverse holds in the other group (Smith, 1776, Schwartz & Hoeksema, 1998). If each group can buy the product it is less suited to manufacture, more cheaply than it can make it, trade benefits both groups, and mutualism can evolve without initial disadvantage to either (Schwartz & Hoeksema, 1998). The exchange of goods or services is as central to natural as to human mutualisms: indeed, David Ricardo’s principle of comparative advantage governs the evolution of mutualism as it does the patterns of local or international trade. Zooxanthellae exchange carbohydrates for their host coral’s nutrients (Muscatine & Porter, 1977; Goreau et al., 1979; Trench, 1987); some plants wrap their seeds in fleshy fruit, so that animals may eat the fruit and drop or defecate the seeds far from the parent and the pests specialized upon its species (Howe & Smallwood, 1982).

Indeed, in nature, such exchanges allow mutualism to evolve much more readily than social behaviour among conspecifics. After all, an animal’s fellow group members are its closest competitors for food, mates and shelter. As Darwin (1859: 75–76) realized, members of different species normally compete less intensely with each other, especially if the species are only distantly related. Indeed, the most familiar mutualisms—corals and zooxanthellae, or plants and pollinators—involve members of different kingdoms pooling contrasting abilities for mutual benefit. A strain of Wolbachia bacteria evolved from parasitism to mutualism with their Drosophila hosts within two decades (Weeks et al., 2007): has social behaviour ever evolved so quickly? Although bacteria have evolved various forms of cooperation (West et al., 2006), mutualisms involving mitochondria and chloroplasts evolved long before metazoans with division of labour among cells. Mutualisms between plants and mycorrhizae (Remy et al., 1994) and between clams and chemo-autotrophic bacteria (Distel, 1998) evolved long before group life among vertebrates or insects. In the evolution of cooperation, does Gause’s (1935) principle of competitive exclusion trump Hamilton’s (1964) kin selection, at least among eukaryotes?

Natural mutualisms divide more readily than human ones into long-term relationships, symbioses (Douglas, 2010), which are often essential to the survival of both participants, and brief exchanges. The former include major evolutionary innovations, such as corals housing symbiotic zooxanthellae, termites and their wood-digesting gut biota, and leaf-cutter ants that cultivate a leaf-digesting fungus. Brief-exchange mutualisms include interactions between plants and their seed-dispersers, and between most animal-pollinated plants and their pollinators. The former are more like long-term business contracts; the latter are more like that between mobile buyers and the shopkeepers competing for their custom.

Long-term mutualisms dissolve if one party ceases to need, or to benefit from, the other’s services. Corals expel zooxanthellae when they can no longer provide carbohydrates (Trench, 1997; Baker, 2001). The saxifrage herb Lithophragma preferentially aborts flowers full of eggs of the flower-parasitizing pollinator moth Greya when cheaper pollinators are available (Thompson & Cunningham, 2002). Interactions between African acacias and their ant defenders become more aggressively antagonistic if large herbivores, whom the ants deter, are removed (Palmer et al., 2008). Live-in nematodes compromise the fitness of their fig-wasp host if the fig fruit their host enters is likely to harbour other fig wasps whose young can be colonized by these nematodes’ young (Herre, 1993). Mutualisms and commensalisms discandy when one party ceases to benefit from, or becomes less dependent on, the other.

Moreover, even a defensible common good seldom leads to instant mutualism. At the dawn of trade the distinction between ‘legitimate trade’ and piracy was almost nonexistent (Bryant, 1953: 116–7, 375–6), as it still was, more recently, in the early stages of colonialism. Nevertheless, if legitimate trade benefits both populations, and piracy can be limited, an effective body of law develops to protect visiting merchants and their hosts from cheaters. The relevance of these economic examples to biology is illustrated by the roles of seed-eating agoutis as seed-dispersers (Smythe, 1989), cone- or flower-parasitizing beetles, moths or wasps acting as pollinators (Pellmyr & Huth, 1994; Schneider et al., 2002; Herre et al., 2008), and the mutualism that developed between aerobically respiring protomitochondria and the cells they entered (Margulis, 1993). One question, therefore, is what circumstances allow a potentially beneficial exchange between members of different species to provoke the evolution of a mutualism sufficiently proof against cheaters to serve the common good of the participants?

Although some (unoriginal) mathematical theory will be used to organize this paper’s approach, the real work will follow the English common law’s emphasis on relevant ‘case histories.’ To wit, I will present an ordered series of examples chosen to elicit evidence for the factors that most vitally influence the likelihood of mutualism, and how they do so.

Behind the question of how mutualisms that resist cheaters can evolve lies another. Smith (1776) wrote as if forms of cooperation that allow a group of people to compete more effectively with others for the means to live were so many steps towards a more mutualistic civilization. Sen (2009) thought that the successive rectifications of different classes of injustice would lead to a reasonably just and mutualistic human society, contrasting his view with Rawls’ (1971) outline of an ideally just society, which included little discussion of the steps by which it could be attained. Wilson (1997) denied that the evolution of such mutualisms would transform ecological communities into commonwealths of interdependence. What role does mutualism play in the evolution of communities? To what extent does the development of successive mutualisms by natural selection represent steps towards transforming ecosystems into genuine commonwealths that, like well-governed civilizations, serves the common good of its members?

Conditions for mutualism: theory

What conditions must apply if mutualism serving the common good of both parties is to evolve? Mutualistic behaviour increases the inclusive fitness of an individual i of species 1 if the benefits to i from members of species 2 outweigh the costs to i of benefiting species 2. Similarly, mutualistic behaviour increases the fitness of an individual j of species 2 if j’s benefits from species 1 outweigh the costs to j of benefiting species 1 (Trivers, 1971). This is more likely if the good each species provides the other costs donors little while greatly benefiting recipients (Schwartz & Hoeksema, 1998; Fletcher & Zwick, 2006).

To be specific, let the inclusive fitness W1i of member i of species 1, and the inclusive fitness W2j of member j of species 2 be


Here, W10 is the inclusive fitness of individual i if it does not cooperate, assumed the same for all members of the species: this is the ‘base fitness.’P1i is the level of cooperation of i’s phenotype, 1 if it cooperates and 0 if not, c1i is the cost of cooperating, P2(i) is the sum of the levels of cooperation of the members of species 2 with which i interacts, and b2 is the benefit to i from each individual that cooperates. Likewise for individual j of species 2 Mutualism spreads (Fletcher & Zwick, 2006) if


In other words, mutualism spreads if


where r21 is the regression of the sum of the levels of cooperation of the members of species 2 interacting with i on i’s level of cooperation, and r12 is the regression of the sum of the levels of cooperation of members of species 1 interacting with j on j’s level of cooperation.

Costly mutualism (c1 > 0, c2 > 0) only evolves if, on the average, an individual i of species 1 benefits from providing goods or services to species 2 and an individual j of species 2 benefits from providing services to species 1. In other words, the bread an individual casts upon the waters must return to it, or its relatives. This cannot happen in the panmictic world of Volterra’s (1931) equations, where all individuals of a cooperator’s species, cooperators and noncooperators alike, benefit equally from the consequences of one cooperator’s good deeds (Trivers, 1971; Wilson, 1980). For cooperation to benefit cooperators, cooperators of one species must be differentially associated with cooperators of the other (Trivers, 1971; Frank, 1994; Foster & Wenseleers, 2006).

Mutualisms without division of labour

The conditions of mutualism derived above apply readily to gregarious fruiting in dipterocarps (Sun et al., 2007). A tree fruits conjointly with trees of other species if fewer of its seeds or seedlings are eaten as a result. A tree only benefits from an episode of gregarious fruiting by bearing fruit when the others do. Gregarious fruiting of dipterocarps benefits the seeds of all participants whose seeds are near enough to those of other participants to help satiate, and to benefit from the satiation of, seed predators: the cooperators involved must be associated to benefit. The same must apply for gregarious reproduction in algae (Clifton, 1997) or corals. In these mutualisms, there is no way to cheat.

Another mutualism with no division of labour is mixed-species bird flocks. Here, several species of understory bird forage together for insects along a regular beat through the forest understory. Birds of each species either eat a distinct set of insects, or forage in a distinct manner. The primary advantage of flocking is not flushing food for each other, but more eyes to watch for predators (Willis, 1972: 140–142; Thiollay & Jullien, 1998; Sridhar et al., 2009). If the alarm call of a bird spotting a predator tells the predator that the bird has seen it and can no longer be caught (cf. Smythe, 1970), the flock’s welfare does not depend on some members taking exceptional risks. Birds may cheat by spending too little time looking for predators. Nonetheless, this mutualism seems unenforced.

Some mutualisms lacking division of labour, however, are more vulnerable to cheating. A Müllerian mimic benefits directly by associating with other poisonous or distasteful species with the same warning coloration, which allows predators to learn to avoid distasteful prey more quickly, reducing the frequency of attacks on distasteful insects (Sherratt, 2008). Their warning coloration is a signal indicating distastefulness. Signals, however, can lie: similar colour patterns do not necessarily imply similar distastefulness. Batesian mimics, edible butterflies mimicking poisonous models, are cheaters, benefiting from others’ distastefulness while reducing the protection distasteful butterflies derive from Müllerian mimicry (Fisher, 1930; Gilbert, 1983). One defence of Müllerian mimics against such cheaters is that cheaters apparently lose their advantage when they become too common (DeVries, 1987: 25).

Mutualisms with division of labour (trade or exchange)

Trade is favoured when the benefit to B of goods provided by A outweighs the cost to B of the goods it provides A in return, and vice versa. The same is obviously true for mutualism between members of different species (Schwartz & Hoeksema, 1998; Fletcher & Zwick, 2006; Foster & Wenseleers, 2006). What keeps species A from cutting costs by stealing from species B? Granted, this strategy is rarely sustainable, but it is an equally unsustainable, though moderately frequent, strategy for cells to turn cancerous, and their descendants to spread, killing the individual to which they belong. More generally, what ensures that an individual of a mutualistic species benefits from providing goods to members of the partner species? Calculating the conditions under which an individual benefits by cooperating is often a surprisingly subtle business (Foster & Wenseleers, 2006). One thing is clear, however: mutualism evolves only if mutualists of one species are differentially associated with mutualists of the other (Fletcher & Zwick, 2006). In what follows, I will discuss representative mutualisms to show how they are maintained, and what (possibly very different) factors caused them to evolve.

Long-term mutualisms (symbioses)

First, I focus on circumstances favouring long-term mutualisms where individuals of at least one partner species spend over half their lives associated with the other. Henceforth, following Douglas (2010: 5–12), I call such long-term mutualisms symbioses. Such mutualisms are usually obligate, in the sense that neither species can live without the other. Usually, the partners are of very different size. Individuals of the larger species, the host, house many individuals, called symbionts, of the smaller species. The host, for whom cheating is often self-defeating, tends to be the dominant partner. Different host species use different ways to keep symbionts from becoming parasites (Douglas, 2010: 70–72, 91), as we shall see. Indeed, symbionts rarely evolve into parasites (Douglas, 2010: 36), although many symbionts attract parasitic third partners.

Vertically transmitted symbionts

Mutualism evolves most readily between hosts and endosymbionts that pass only from their host to its offspring (Herre et al., 1999). Endosymbiosis usually ensures that when cooperators of both species become associated they remain associated. Here, the host benefits by helping beneficial endosymbionts, whereas the endosymbionts’ offspring benefit when endosymbionts benefit their host. The most spectacular of these mutualisms is that between eukaryotic cells and their organelles—mitochondria and chloroplasts. Perhaps two billion years ago, the ancestors of mitochondria were free-living aerobic bacteria which were soon to parasitize, or survive engulfment by, anaerobic archaean hosts (Margulis, 1993), thereby associating the future partners in mutualism. The guests’ aerobic respiration complemented their hosts’ food-getting ability, so the guests became mutualists. Eventually, the guests became so dependent on their hosts that they passed only from parent to offspring. Thus their fitness reflected that of their hosts, so the guests were subject to a ferocious group selection in their host cells’ service (Leigh, 1983, 1991, 2010). The mathematics are very clear (Hamilton, 1975; Leigh, 1983): what drives this group selection is the inbreeding among each cell’s symbionts, which reduces their within-cell genetic variance, depriving within-cell selection of the traction needed to oppose group selection. Despite this powerful group selection, however, harmony is not complete. To avoid destructive conflicts, ‘civil wars,’ between organelles from different parents, sexually produced zygotes normally inherit organelles from only one parent, usually the mother. As a result, organelle genes benefit from all-female sex ratios. Revealing a trace of ancient piracy, organellar mutants sometimes impose all-female sex ratios on their hosts, although these mutants are usually suppressed later by mutants in the host’s nuclear genome (Eberhard, 1980; Cosmides & Tooby, 1981).

Bacterial endosymbionts of various sap-sucking insects are passed on so exclusively from parent to offspring that they cospeciate with their hosts (Moran & Telang, 1998). Did these endosymbionts begin as parasites, undigested prey or commensals? These bacteria usually synthesize vitamins, amino acids, and other compounds that their hosts need but cannot obtain from their food. Now, the bacteria cannot live without their hosts who, in turn, cannot live without their bacteria. Some of these associations are very old: the common ancestor of Sulcia, a genus of endosymbionts living in plant-hoppers, leaf-hoppers, cicadas and spittlebugs, lived in their common ancestors 290 million years ago. The best known of these endosymbiotic bacteria is Buchnera aphidicola. Their hosts, pea aphids (Acyrtosiphon pisum), that live on nutrient-scarce fluids that they suck from xylem or phloem, need these Buchnera to synthesize vitamins and amino acids these hosts cannot live without. The genome of Buchnera aphidicola is an 0.64 MB subset of the > 2 MB genome of Escherichia coli K-12. Buchnera’s metabolic network programs only 27% of E. coli’s metabolic reactions. Unlike E. coli, Buchnera’s metabolic network has almost no redundancy. About a third of the reactions in Buchnera’s metabolic network are needed to synthesize amino acids its hosts need. Buchnera’s metabolic network has been modified so that, to synthesize the purines these symbionts need to replicate, they must synthesize as a precursor the histidine their host needs to live. Buchnera and their hosts are utterly interdependent (Thomas et al., 2009).

Symbiosis by strict vertical transmission has its drawbacks: the symbionts are severely inbred. Mutants in these symbionts frequently disable genes, which are then dropped from the genome, which consequently shrinks. Often, genes for genetic repair are disabled, and mutation rates increase sharply (Douglas, 2010: 79–81). As in Buchnera, endosymbiont genomes are often far less than half the size of the genomes of their free-living ancestors. Endosymbiotic gut bacteria are usually housed within a special membrane called a bacteriosome. Bacteriosomes may carry out reactions, which endosymbionts need, but have lost the genes for (Moran et al., 2008). Deterioration of the genome may explain why symbionts are occasionally replaced by invaders whose genomes are more complete (Douglas, 2010: 44). Many genes of mitochondria and chloroplasts have been transferred to the nuclear genome, where they are better protected from the effects of genetic drift and mutation pressure (Douglas, 2010: 82).

Other endosymbionts that normally pass from parent to offspring are far less dependent on their hosts (Moran & Telang, 1998). When its genome harbours a temperate lambda-bacteriophage, the bacterium Hamiltonella defensa kills the eggs laid in the pea aphid Acyrtosiphon pisum by parasitoid wasps, or the larvae hatching from them. Hamiltonella only benefits its host by killing parasitoids: if these wasps are absent, Hamiltonella are useless to their hosts. Hamiltonella can disappear from host lineages and reinvade when parasitoids reappear. Hamiltonella has genes used by pathogens such as its relative, the black plague bacillus, to evade suppression by host immune systems. Hamiltonella’s genome is three times larger than Buchnera’s and acquires new genes as well as losing old ones (Degnan et al., 2009).

In two sets of fungus-farming insects, colonists take the symbiotic fungus from the parental nest and ‘plant’ it when founding a new nest after dispersing. Wood-boring ‘ambrosia beetles’ bore into tree trunks or branches and tunnel in the xylem. They inoculate the walls of these tunnels with a fungus culture they brought from their parents. The fungi digest the wood, beetles eat the fungi. Neither can live or reproduce without the other (Mueller et al., 2005). This symbiosis arose at least seven times in the last 60 million years. It evolved so easily because, to disperse from tree to tree, ancestors of the fungi rode beetles.

Although these beetles live in small family groups, some species enter into mutualism with antibiotic-producing bacteria to protect their fungus from devastating infection. Southern pine beetles, Dendroctonus frontalis, depend on the fungus Entomocorticium sp. to transform pine wood into suitable food for their larvae. The beetles carry this fungus from the parental nest to start their own. The fungus, however, can be replaced by a fungus, Ophiostoma minus, that rots wood more quickly without providing food suitable for the beetles. To counter this threat, the beetles culture a bacterium related to Streptomyces, which, like the fungus it protects, is passed on from parent to offspring beetles. This bacterium secretes an antibiotic that shuts down Ophiostoma with minimum harm to the Entomocorticium that the beetles need (Scott et al., 2008).

About 50 million years ago, ancestral fungus-growing ants domesticated fungi which they fed with insect frass or corpses, or dead vegetable matter (Mueller et al., 2005). Fungus-growing ants are monophyletic: they had a single origin (Mueller et al., 2001; Schultz & Brady, 2008). Fungus-growing ants presumably originated from fungus-eating ants, a habit that associated the future mutualists, when the ants began cultivating the fungus they were eating. Early on, fungus-growing colonies usually numbered less than 100 workers apiece. More recently, fungus-growers gave rise to leaf-cutter ants, which feed fragments of cut leaves to their fungus. Several species of leaf-cutter ants, Atta, develop colonies of millions of workers apiece, with intricate division of labour. In some tropical forests, Atta are the primary consumers of foliage (Hölldobler & Wilson, 2009).

In leaf-cutter ant nests, the fungus they cultivate uses the ants to help maintain the fungal strain’s own genetic uniformity. The fungal strains of different leaf-cutter ant colonies, even if they are of the same species, differ genetically, and make different self-recognition compounds. These compounds cause their fungus to repel counterparts from different colonies, with different compounds, should they come in contact. Leaf-cutter ants fertilize the fungi they cultivate with faecal pellets containing the self-recognition compounds of their fungus, thereby helping suppress ant-fungi accidentally introduced from other leaf-cutter colonies (Poulsen & Boomsma, 2005).

In leaf-cutters, parent–offspring transmission of the fungus is sufficiently reliable to ensure that leaf-cutter ants and their fungus never cheat each other. The mutualism, however, faces a different challenge. Defending themselves and their fungus against pathogens consumes a major part, about 40%, of a leaf-cutter colony’s energy budget (Fernández-Marin et al., 2009: 2267). Soon after fungus-growing ants evolved, a parasitic fungus, Escovopsis, appeared. The symbiotic fungus of each species of fungus-growing ant is susceptible to potentially lethal infection by some species of Escovopsis. To defend themselves against Escovopsis, species with smaller colonies cultivate antibiotic-producing bacteria, Pseudonocardia, whose antibiotics depress, suppress or reverse the growth of infecting Escovopsis. Large Atta colonies defend themselves with antimicrobial secretions from the metapleural glands of the ants themselves. The public health measures of large leaf-cutter colonies is a fascinating business, which testifies to the danger infection poses their symbiotic fungus (Fernández-Marin et al., 2009).

Horizontally transmitted symbionts and partner fidelity

How are mutualisms maintained when one partner is not passed on by the host to its offspring, but must be attracted by the host from its surroundings? Mutualism must serve the partners’ common good. Under horizontal transmission, mutualism seems out of the question if the potential partners’ interests are as irreconcilable as those of human beings and diseases transmitted by insects (Douglas, 2010: 29). Even when mutual benefit is possible, cheating is a danger. Three approaches may restrain cheating, ‘partner fidelity,’‘partner choice’ and ‘partner sanctions’ (Sachs et al., 2004).

Partner fidelity links the fates of interacting individuals of two species involved in mutualism, just as the fates of two people in a rowboat on a stormy sea are linked if both must row for either to survive. In these cases, partners in mutualism share the same fate (Wilson, 1997): injuring one’s partner injures oneself. The common interest among such mutualists can be as absolute as if transmission were vertical.

Partner fidelity arising from association between interdependent partners for much of their lifetimes creates a common interest among the partners in each other’s welfare. Partner fidelity maintains mutualism between three species of branching sponges on Caribbean coral reefs (Wulff, 1997). Sponges of these three species grow faster and survive better when at least some of their branches adhere to—become stuck to—branches of another of these three species. By sticking together, sponges of two mutualist species pool different abilities that enhance the growth and survival of both parties. For example, each species is immune to the diseases of the other two. One of these three species is avoided by starfish. Fragments of the second survive much the best when storms break up these sponges. The third species is best at surviving hurricanes, avoiding breakage by storms and coping with smothering by sediment. On the average, a sponge survives better, at least as fragments capable of regrowth, when attached to a sponge of another of these three species, thanks to their partner’s superior ability to resist certain hazards (Wulff, 1997). A fourth species of branching sponge, however, parasitizes the other three. The parasite’s hit-and-run life history circumvents the demands of partner fidelity by growing and fragmenting far more rapidly so the parasites can export fragments to other partners before ruining their current one. Sponges of the parasite species grow faster and survive better when adhering to a sponge of one of the mutualist species, thanks largely to the mutualists’ superior strength. Sponges of the mutualist species, however, grow more slowly and die faster when adhered to by the parasite (Wulff, 2008). Only the parasite’s hit-and-run strategy lets it escape the consequences of the harm it inflicts on its partners.

Many species of plants have specialized structures that house and feed ants. In turn, these ants protect their host plants from herbivores and encroaching vines (Heil & McKey, 2003). A famous example is the mutualism between bull-thorn acacias, Acacia cornigera, and the plant-protecting ants, Pseudomyrmex ferruginea housed by these acacias in their hollow thorns. Janzen (1966) showed that each species depended on association with the other. This mutualism, too, is maintained by partner fidelity. Colonizing Pseudomyrmex ferruginea take two years to produce their first reproductives, whereas a year without ant protection is often enough to kill the host tree (Janzen, 1975). Benefiting the current host is by far the best way to ensure the colony’s future reproduction. Colonies of Pseudomyrmex nigropilosa, however, produce their first reproductives in two months. These cheater ants colonize ant-acacias without protecting them: they can spread to new hosts before their cheating seriously compromises their current one (Janzen, 1975). For live-in ants, long life cycles are a form of partner fidelity that promotes mutualism with the acacias (Wulff, 2008).

A striking mutualism maintained by partner fidelity involves a third category of fungus-farming insects: fungus-farming termites (Aanen et al., 2002; Mueller & Gerardo, 2002; Mueller et al., 2005). These termites grow a fungus, Termitomyces, on the faeces of workers that have fed outside the nest on wood, dry grass or leaf litter. The termites eat the nutritious fungus and deposit its indigestible asexual spores on the ‘fungus comb,’ renewing the fungus culture. Fungus-farming termites evolved 30 million years ago in east African rainforest. Farming fungus makes these termites major decomposers in the palaeotropics. Their societies can be impressive: a nest may have millions of workers in a volume of several cubic metres (Aanen et al., 2002, 2009).

In most fungus-farming termite species, nest founders acquire their fungus from the abundant sexual spores near the nests they are founding. These spores are produced by mushrooms formed by fungus cultures of established termite nests. Thus, this fungus is transmitted horizontally. Each colony’s fungus is genetically unique, and genetically uniform (Aanen et al., 2009). Fungal hyphae fuse only with clonemates. Fusion among hyphae disproportionately increases their production of asexual spores, giving overwhelming advantage to the most common clone, whose hyphae are most likely to find partners to fuse with. Even though a new termite colony starts with a genetically diverse fungus culture, this frequency-dependent selection soon allows one clone to take over the culture. This clone’s fate is identified with its host colony’s. Partner fidelity works: only two small clades of fungus-farming termites have evolved means to transmit the fungus directly from parent to offspring (Aanen et al., 2009).

Horizontally transmitted symbionts and partner choice

In most horizontally transmitted mutualisms, at least one party has some chance of changing to another partner: partner fidelity is not complete. In Herre’s (1993) study of nematodes that live in fig-pollinating wasps, the prospect that a nematode’s young could find homes in the young of a fig wasp different from its mother’s let selection transform commensals into overt parasites. The common interest among the partners of such mutualisms as corals and their zooxanthellae or trees and their mycorrhizae, however, is strong enough to allow these mutualisms to persist despite opportunities for switching partners. How can the common interest among mutualists, the benefit of pooling complementary abilities for the common good, be defended against cheaters whose fate does not depend utterly on their partners’ success?

A striking mutualism maintained primarily by partner choice involves the 5 cm Hawaiian bobtail squid, Euprymna scolopes (see Moynihan, 1983) and its symbiotic bioluminescent bacteria Vibrio fischeri (McFall-Ngai & Ruby, 1998). The luminescent bacteria adjust the shade of the squid’s silhouette to match the shade imparted by downwelling moonlight to the surrounding water (Jones & Nishiguchi, 2004), presumably making the squid harder for predators or prey to see. As 2 mm hatchlings less than a day old, these squid attract their Vibrio from the water, where Vibrio represent one ten-thousandth of the million bacteria per cubic centimetre of seawater (Nyholm & McFall-Ngai, 2004). Partner fidelity is not complete because each squid expels some symbionts alive into the water every morning, some of which find another partner. The first question is one of partner choice: how do these baby squid attract the right bacteria and exclude all the others?

To find their new home inside the squid, Vibrio fischeri must run an obstacle course designed to exclude both defectives of their own species and bacteria of all other species. This course is quite as rigorous as any that a swain in a traditional fairy tale must run to win his princess. As in a proper courtship, establishing mutualism is the culmination of the right fulfilment of a ritual sequence of signals and acts exchanged between a squid and its Vibrio.

To establish this symbiosis, the squid must first attract Vibrio. The first step to attracting symbionts happens two hours after hatching when the beating of cilia on the squid’s underside, around its light organ, create aggregations of bacteria around the pores leading to the crypts where the squid houses its symbionts. The presence of any kind of bacteria causes the squid to produce mucus near these pores. Live gram-negative bacteria of many kinds enter this mucus. Within a few hours, however, Vibrio fischeri are the only gram-negative bacteria in the mucus. A few hours after taking over the mucus, the Vibrio start to move through the pores, down the ducts and into the crypts (Nyholm et al., 2000; Nyholm & McFall-Ngai, 2004).

To reach these crypts and colonize them successfully, however, the Vibrio need several abilities, many of them uncommon. First, they need to move: their flagella must be functional. Next, they must be able to survive high concentrations of nitric oxide, NO, in the ducts. Signals from the Vibrio, however, cause the squid to lower the concentration of nitric oxide in the ducts and crypts by 80% in eight hours. Next, in the crypts the Vibrio encounter an enzyme, halide peroxidase, that catalyses hypochlorous acid, a potent microbicide, from hydrogen peroxide and chloride ions. Hypochlorous acid serves as the first line of anti-microbial defence in mammalian immune systems. The Vibrio parry this threat by producing toxins like those of their relatives, Vibrio cholerae, that reduce the abundance of halide peroxidase. Some Vibrio fischeri enzymes, moreover, use hydrogen peroxide, diminishing the supply needed by the halide peroxidase enzyme to catalyse its microbicide. Finally, to establish viable symbiosis, these Vibrio need working bioluminescence genes and functional regulators which induce bright luminescence when the concentration of Vibrio reaches a certain threshold. Bioluminescence consumes oxygen. In the crypts, luminescent Vibrio outcompete lightless counterparts, perhaps because the oxygen consumption demanded by luminescence reduces the danger to luminescent Vibrio from oxidants (reactive oxygen species) (McFall-Ngai, 1999; Visick et al., 2000; Nyholm & McFall-Ngai, 2004).

The crypts provide a hospitable home for Vibrio fischeri. These bacteria double in number every half hour until they fill their crypts. Once the Vibrio reach the crypts and begin to multiply, they release tracheal cytotoxin, a fragment of the peptidoglycan molecules on their surface, and lipopolysaccharides. Tracheal cytotoxin causes extensive tissue damage in whooping cough infections: this compound reflects these Vibrio’s piratical ancestry (Koropatnick et al., 2004). In squid, tracheal cytotoxin and lipopolysaccharides induce cell death in the organs that first called the Vibrio to their host. These and other signals from the Vibrio colonists cause the ducts to narrow within a few hours. Within five days, their tracheal toxin and lipopolysaccharides have caused the irreversible disappearance of the fields of cilia and the mucus-secreting structures that first called the Vibrio to the squid. These compounds also induce changes in the crypts that make them more effective light organs (Koropatnick et al., 2004). The squid and its Vibrio are now committed to each other, even though the squid expels 95% of their Vibrio into the water every morning. If the most luminescent bacteria are held most closely to the crypts’ walls, these daily purges help enforce the mutualism (Sachs et al., 2004). These expulsions also insure that baby squid have symbionts to attract from the surrounding water (Nyholm & McFall-Ngai, 2004). Given the piratical nature of these Vibrio’s relatives, and their own potential for piracy, it is no wonder that the squid subject them to ferocious probation before housing them and take severe measures to ‘keep them honest’.

Horizontally transmitted symbionts and partner sanctions

Now we ask: how might partner sanctions enforce mutualisms? Partner choice sometimes selects serviceable mutualists, as when bobtail squid test their symbionts for serviceability before accepting them. It is often difficult, however, to detect and reject useless or harmful ‘applicants.’ Mutualism is more often maintained by ‘punishing’ nonperforming partners.

A mutualism that depends on both partner choice and partner sanctions is the 60-million-year-old relationship between many species in the plant family Leguminosae and the ‘nitrogen-fixing’ rhizobial bacteria they house in nodules on their roots (Kiers & Denison, 2008). The plant supplies its bacteria with carbohydrates and various mineral nutrients; in return the bacteria supply the plant with ammonia, which it synthesizes from atmospheric nitrogen, N2, in a remarkably energy-consuming process.

Mutualism initiates between a plant and its rhizobial bacteria in the soil around its roots through a complex exchange of chemical signals—a ritualized conversation in chemical code, as it were (Fisher & Long, 1992; Van Rhijn & Vanderleyden, 1995). The rhizobial bacteria take advantage of some pre-existing responses of plants to another major mutualist, arbuscular mycorrhizal fungi (Markmann & Parniske, 2008). In a typical case, rhizobial bacteria are attracted by particular exudates from roots of an appropriate host. If a bacterium’s cell surface proteins match lectins on the root surface, the bacteria attach to a root. In response to appropriate compounds from the plant, bacteria of the right type release ‘Nod factors,’ nodulation factors, which cause root hairs to branch and curl, and cause the creation of a pocket in the cell wall where the root hairs trap these bacteria. The plant then forms an ‘infection thread,’ a tube that grows inward from the pocket. This ‘thread’ causes the initiation of root nodules to house the bacteria and provide the route by which the bacteria enter their new homes (Van Rhijn & Vanderleyden, 1995).

This conversation differs from the exchanges between baby squid and colonizing Vibrio: the plants and rhizobial bacteria merely exchange signals, and bacterial signals can lie. Colonizing bacteria do not face the comprehensive testing vibrios must survive to colonize a squid. The contrast is a bit like that between the accomplishments demanded of a male rat, and a male bowerbird, to secure a mate. The achievements demanded of a male bowerbird (Marshall, 1954; Diamond, 1988), like those demanded of a squid’s Vibrio, cannot be faked. On the other hand, because a plant cannot screen ‘applicant’ rhizobia properly, the worse colonize along with the better, and the plant must enforce cooperation by punishing nonperformers (Kiers & Denison, 2008). True, it is self-defeating for a plant not to feed bacteria that provide it with ammonia: starved bacteria cannot make ammonia. On the other hand, bacteria would profit by not devoting large amounts of energy to synthesizing ammonia. Kiers et al. (2003) found that if a nodule’s bacteria are prevented from synthesizing ammonia, the plant starves them of oxygen, which halves their fitness, thus enforcing mutualism.

The symbiosis between corals and their zooxanthellae has transformed ecosystems. Symbioses between corals and zooxanthellae underlies the success of scleractinian reef-building corals from the Triassic onward (Stanley, 2006); symbioses between algae and reef-building tabulate corals may have driven the enormous expansion of late Devonian reefs (Talent, 1988: 340–344). Yet this symbiosis is far less understood than that between legumes and nitrogen-fixing bacteria. Coral zooxanthellae belong to the dinoflagellate genus Symbiodinium, whose members differ genetically as much as do other dinoflagellates of different orders. Some corals transmit their zooxanthellae directly to their descendants (Trench, 1987). The supply of carbohydrates by symbionts to corals and the supply of nutrients by corals to symbionts, however, provide the foundations for a mutualism that survives horizontal transmission. Like the rhizobium-legume symbiosis, horizontally transmitted coral/zooxanthellae symbioses are maintained by a combination of partner choice and partner sanctions.

Too little work has been carried out on how corals ‘call’ their zooxanthellae. Unlike the squid symbiont Vibrio fischeri, free-living members of symbiotic species of Symbiodinium have rarely been observed in nature (Baker, 2003). Motile symbionts, however, establish symbiosis with coral hosts in the laboratory (Trench, 1987, 1997). Koike et al. (2004) studied one stage of symbiosis formation in the octocoral Sinularia lochmodes. This octocoral produces a lectin which binds to the cell surfaces of laboratory-cultured, motile Symbiodinium. This lectin transforms these motile cells into the ‘coccoid’ nonmotile form assumed by Symbiodinium in coral polyps. This lectin’s effectiveness in turning motile into nonmotile forms differed for different strains of Symbiodinium. It had no such immobilizing effect on nonsymbiotic species of dinoflagellates. Indeed, the lectin killed two of the three nonsymbiotic species tested (Koike et al., 2004). How this octocoral calls its Symbiodinium in close enough for its lectin to affect them is unknown. How many other corals use such lectins (so like in effect to the lectins that help mediate the legume–rhizobium symbiosis) to help choose appropriate symbionts and exclude inappropriate ones is likewise unknown. Nonetheless, Koike et al. (2004) have established a suggestive beginning to the study of partner choice in corals.

Corals expel their zooxanthellae when they no longer perform adequately (Baker, 2001). Corals can call in other symbionts after expelling their predecessors (Lewis & Coffroth, 2004). Symbionts of different strains function best at different depths and differ in their tolerance of temperature variation (Baker, 2003; Iglesias-Prieto et al., 2004; Stat et al., 2008). Indeed, horizontal transmission and the ability of corals to replace their symbionts enables these corals to adapt better to changing environments (Baker et al., 2004; Rowan, 2004). This adaptability is limited, however. Past instances of global warming enabled modified pearl oysters, Lithiotes, to temporarily replace symbiotic scleractinian corals in the early Jurassic, and other bivalves, rudists, to repeat the feat in the late Cretaceous (Fraser et al., 2004).

Early flowering plants were fast-growing weeds of disturbed streamsides (Wing & Boucher, 1998; Royer et al., 2010), which employed animals as pollinators (Grimaldi & Engel, 2005: 616–622; Hu et al., 2008). Over 100 million years ago, some plants evolved specialized flowers that attracted particular pollinators such as bees (Poinar & Danforth, 2006) that would seek out flowers on conspecific plants (Crepet, 1984; Crepet & Nixon, 1998; Gandolfo et al., 2004). Such dedicated pollinators maintain genetic variation even in rare plant species (Grimaldi & Engel, 2005: 618). This variation allowed a plant species to survive when kept so rare by specialist pests that many of its plants escaped their attention. In such a species, plants could therefore devote more resources to fast growth rather than anti-herbivore defence (Coley et al., 1985). Domesticating dedicated pollinators triggered extensive diversification in both plants and pollinating insects (Crepet, 1984; Wing & Boucher, 1998; Grimaldi & Engel, 2005: 460). Eventually, a diverse set of faster-growing, animal-pollinated, flowering plants replaced a far less diverse array of better-defended, slower-growing wind-pollinated gymnosperms, thereby transforming tropical forest ecosystems (Corner, 1964; Regal, 1977; Leigh, 1999; Leigh et al., 2004). Although most animal pollination involves mutualisms of brief exchange, some obligate pollinator mutualisms are enforced by partner sanctions.

A simple example of an obligate pollinator mutualism involves yuccas, Yucca spp., and the moths, Tegeticula spp., that pollinate them (Pellmyr, 2003). Larvae of these moths live only on yucca seeds. When a female moth emerges from its pupa, it gathers pollen from its hosts’ flowers. The moth then flies off and lays eggs in the flowers of several plants of its favoured species of Yucca. It pollinates every flower in which it tries to lay eggs, to provide its young with seeds to eat. These moths are the yucca plants’ only pollinators: the plants cannot reproduce without their help. The pollinating moths descend from seed predators. These moths would benefit by laying enough eggs that their young eat all their hosts’ seeds. They do not succeed in this ultimately self-defeating end. First, it is difficult for a moth to lay eggs in the correct section of a flower’s ovule. Second, yuccas always abort a large proportion of their flowers, including pollinated ones, so yucca plants increase their fitness by choosing to abort both inadequately pollinated flowers, and flowers with too many moth eggs. The behaviour of yucca plants selects for moths that pollinate effectively but lay too few eggs to compromise their hosts’ reproduction (Pellmyr & Huth, 1994). A similar mutualism involves Glochidion acuminatum (Phyllanthaceae) and its seed-eating obligate pollinator, the moth Epicephala sp. (Goto et al., 2010).

Although yuccas now need Tegeticula to pollinate them, these moths’ ancestors were seed predators. Because most insect seed predators specialize on seeds of a single species (Janzen, 1980), the tendency of these moths to lay eggs in flowers with pollen made it possible for them to become faithful pollinators. The potential for piracy, however, has not disappeared completely. Cheater species of Tegeticula have evolved, that lay eggs in yucca flowers when it is too late for the plants to abort them, and when pollinating them ceases to make a difference. A cheater moth cheats a species of Yucca different from those pollinated by its ancestors: cheaters arise as a result of a ‘host shift’ (Pellmyr et al., 1996).

The best-known obligate pollinator mutualism involves figs and their pollinator wasps (Corner, 1952; Janzen, 1979; Herre et al., 2008). Each species of fig has one or more dedicated species of pollinating wasp (Wiebes, 1979; Molbo et al., 2003). These wasps are extraordinarily effective pollinators. Although a mature pollinating wasp lives three days or less, pollinator wasps of monoecious figs can carry pollen from one tree to another 10–100 km away (Nason et al., 1998; Harrison & Rasplus, 2006). Ripe fig fruit also attract bats and other animals that disperse their seeds long distances (Kalko et al., 1996). Although monoecious fig species are often very rare, with one adult per 10 or 100 ha, they maintain extraordinary genetic variation, thanks to their wasps (Nason et al., 1998; Harrison & Rasplus, 2006). If employing animals as pollinators and seed-dispersers allowed flowering plants to replace gymnosperms, then figs, with their long-distance pollinators and seed-dispersers, must be the ultimate flowering plants. Indeed, fig trees are notorious for fast growth (Janzen, 1979): a Panamanian Ficus insipida had the highest photosynthetic capacity of any tree yet measured, and growth to match (Zotz et al., 1995). The organization of fig trees for growth, not longevity, has considerable impact on their forest: ‘the abundance of fig species is a good measure of the richness of the environment in plant and animal life. By leaf, fruit and easily rotted wood, fig-plants supply an abundance of surplus produce’ (Corner, 1967: 24).

This wondrous pollination system has many costs. At any time of year, each fig species must have some trees ready to receive wasps, for released wasps must have some place to go (Janzen, 1979; Windsor et al., 1989). Some trees must accordingly fruit at seasons when seedling prospects are poor. Larger fruit attract larger seed-dispersers, which fly further, carrying their seeds further from their parents and their pests (Seidler & Plotkin, 2006). A large syconium, however, needs more foundresses to pollinate its flowers. The more foundresses, however, the larger the proportion of males among their young (Herre, 1985): these males do their tree no service. Moreover, a large syconium heats up more than a small one more readily cooled by movement of the surrounding air. Excess heat kills wasps. Larger fruit must therefore transpire more water per unit weight to keep their wasps from overheating (Patiño et al., 1994). Despite these costs, however, fig-pollinating wasps have propelled their figs into an extraordinary variety of habitats and ways of life, and have allowed them to evolve remarkable diversity despite their slow pace of speciation (Corner, 1967; Janzen, 1979; Machado et al., 2001).

Like yucca moths, fig-pollinating wasps evolved from seed predators. The relatives from which they diverged 90 million years ago, when the mutualism began, still prey on fig seeds (Machado et al., 1996, 2001; West et al., 1996). At least two wasp species descended from pollinators have reverted to seed predation, one without colonizing a new species of fig (Herre et al., 2008). The benefits, costs and ecosystem impacts of this mutualism far exceed those of, say, yuccas and yucca moths (Janzen, 1979).

In ancestral fig trees and all neotropical ones, a fig fruit or ‘syconium’ starts as a flowerhead turned outside in—a perforated ball lined on the inside by flowers. When ready for pollination, the tree’s syconia exude a scent, characteristic of its species, that attracts appropriate wasps from the moving air above the canopy. One or more mated, pollen-bearing, female wasps, the ‘foundresses,’ enter a syconium, pollinate its flowers, and lay eggs in 50–60% of their ovules. The average number of foundresses per pollinated syconium varies from 1 to 5, according to the fig species (Herre, 1987, 1989). Each wasp larva matures within a single ovule. When the syconium’s wasps eclose, about a month later (Janzen, 1979), they mate with each other, the males die, and the mated females, now dusted with pollen, leave their syconium. A tree’s syconia develop in synchrony, so they all call wasps within less than a week, and they release their young during the same few days a month later. These wasps must therefore find another tree with syconia ready to pollinate and lay eggs in (Janzen, 1979; Herre et al., 2008). The syconia then ripen and become ready for animals to eat them and disperse their seeds.

There are two ways wasps might benefit by cheating their figs. They could spare the time and trouble of pollinating their flowers, for the ovules where they lay eggs develop anyway (Jandér & Herre, 2010). They could also lay eggs in every flower, annihilating the tree’s reproduction. Some fig species solve the first problem by producing such abundant pollen that emerging wasps are so dusted with pollen that they cannot avoid pollinating the flowers in the syconia where they lay eggs. The earliest figs did this. Other species make far less pollen, depend on their wasps to actively gather pollen from their natal syconia, and actively pollinate flowers in the syconia where they lay eggs. In these fig species, some wasps cheat. Syconia that are not pollinated are often aborted, and their wasps’ fitness reduced in other ways. Fig species with stronger sanctions have fewer cheaters among their wasps (Jandér & Herre, 2010). Neotropical figs apparently solve the second problem by making 40–50% of their ovules wasp-proof. Learning how they do so is a work in progress by Ellen Suurmeyer and is no part of this story. Some paleotropical fig species solve their problem another way. Some of their trees are ‘gall-figs’ where every flower’s ovule is intended to house a wasp larva. The other trees are ‘seed-figs.’ How wasps are enticed into pollinating seed-figs, thereby in effect sterilizing themselves, is an open question (Patel et al., 1995; Anstett et al., 1997). In some of these fig species, pollinating wasps tend to avoid this dead-end job (Anstett et al., 1998).

Mutualisms of brief exchange

Mutualisms of brief exchange have played an essential role in making ecosystems more mutualistic, as we shall see. Such mutualisms are perhaps the closest analogues to human market economies (Noë & Hammerstein, 1995). Some of these mutualisms represent a sort of reciprocal altruism (Trivers, 1971). Whence comes the repeated interactions between specific cooperators or specific families of cooperators, that allow reciprocal altruism to evolve? How do other brief-exchange mutualisms persist?

Cleaner fish and their clients

Some brief-exchange mutualists have several ways of controlling cheating. One such mutualism is that between ‘cleaner fish,’Labroides dimidiatus, and their client fish from which the cleaners remove ectoparasites, which they eat. Client fish need to be cleaned fairly often. If a patch reef’s cleaners are removed, its nomadic fish soon depart to reefs with cleaners, and the condition of the remaining territorial fish deteriorates rapidly (Trivers, 1971). Although the cleaners live primarily on the ectoparasites they pluck from their clients, they often cheat a bit, taking occasional nips from their clients’ flesh, which they prefer to the ectoparasites (Douglas, 2010: 65). Like a shopkeeper, a Labroides dimidiatus occupies a particular spot, where it is visited by a clientele of loyal customers. Good service attracts customers that keep coming back, so these deeds redound to their perpetrators’ benefit in a sort of reciprocal altruism (Trivers, 1971). Reputation matters to these Labroides: they refrain from nipping at clients if other potential clients are watching (Bshary & Grutter, 2006). If a cleaner cheats egregiously, it confronts partner choice or partner sanctions. Its client either punishes the cheater by chasing it, or moves off to find a more cooperative cleaner (Bshary & Grutter, 2005).

Poorly policed brief exchanges

The most famous mutualisms of brief exchange, however, seem less readily policed. Many plants offer attractive bribes to secure needed services from animals. The bribe is designed so that, to take it, an animal cannot help performing the needed service, such as pollinating the plant’s flowers or dispersing its seeds. The animal has many plants to choose from, whereas a plant may have many animals, of several potentially useful kinds, ‘within call.’ Although competing for the most serviceable partners for long-term mutualism can resemble a ‘job fair’ or ‘marriage market,’ the world of brief exchanges is the closest natural analogue to a big city’s marketplace, where storekeepers compete for the most profitable customers, who in turn seek the best offers (Noë & Hammerstein, 1995). This competition usually leads to a ‘niche differentiation’ where different ‘merchants’ attract different groups of ‘customers.’ In this world, an animal may have only one interaction with a particular plant. There seems to be little room here for partner sanctions or the private vengeance people without access to public law enforcement use to enforce ‘justice’ (Hyams, 2003). Yet the evolution of angiosperms where animals pollinate flowers and disperse seeds in a series of brief mutualistic exchanges transformed tropical forest ecosystems (Corner, 1964; Regal, 1977; Leigh, 1999). What factors allowed cooperation for their common good between plants and the animals that drink their nectar, eat their pollen, or eat their fruit to prevail in this seemingly lawless world?

Plants and their pollinators

Mutualisms of brief exchanges with pollinators are crucial to flowering plants, as we have seen. Yet these mutualisms arose from parasitisms, as we shall see, and the threat that they will degrade again into parasitisms is never far away. The sporophytes of plants that colonized the land produced wind-dispersed spores, which were the first-recorded plant material eaten by terrestrial insects (Labandeira, 2006a: 63). Natural selection transformed the spores of some descendants of these early land plants into pollen. Some pollinators must have descended from such pollen thieves (Crepet et al., 1991): in seeking out plants of species with particularly tasty pollen, pollen of one plant was sometimes carried accidentally to where it would fertilize a conspecific. Another starting point for the evolution of pollinators starts from the ‘pollination drops’ gymnosperm ovules exude at their apical openings. These drops snare airborne pollen belonging to the ovule’s species. The pollination drops of some species are nutritious and attract some insects. Again, searching for plants with particularly tasty pollination drops must have led to accidental pollination. Pollination drops eased the evolution of insect pollination in at least five gymnospermous groups—seed ferns, Gnetales, Bennettitales, conifers of the extinct family Cheirolepidaceae, and cycads (Labandeira et al., 2007). Some of these gymnosperms were probably pollinated by long-proboscid scorpionflies (Ren et al., 2009). Flowering plants must have evolved insect pollination in an analogous manner.

Insect pollination could promote the rapid diversification of flowering plants because the ancestral flowering plants were weeds which could grow fast. The xylem of these weedy plants offered relatively little resistance to the flow of water from roots to leaves, thereby allowing rapid photosynthesis and fast growth. Because early flowering plants could grow fast, employing animal pollinators enabled them to transform tropical forest ecosystems. The evolution of insect pollination in slow-growing plants, such as cycads, had no such effect. In cycads, for example, the ineffective vascular system severely limits photosynthetic capacity and growth rate by restricting the rate at which water can move from roots to leaves (Sperry et al., 2006).

Several questions, however, still need answers. How do plants choose appropriate pollinators? How do plants persuade pollinators to complete their task? How much scope is there for cheating among either plants or pollinators? How does cheating affect the evolution of plants and pollinators?

Different flowers attract different pollinators (Howe & Westley, 1988; Fenster et al., 2004). A flower that attracts bats differs greatly from one that attracts sphinx moths. Plants ‘choose’ their pollinators. More precisely, the plants best at attracting those pollinators that contribute most to their reproductive success are better represented in future generations (Fenster et al., 2004). Some orchids choose long-proboscis sphinx moths as pollinators by storing their nectar in spurs so deep that no other pollinator can reach it. Others attract male orchid bees (Euglossini) with fragrances these bees, but no other animals, gather to use in attracting females (Eltz et al., 2007). Indeed, the origin of a new species often involves ‘choosing’ a new pollinator. For example, in a newly invaded habitat, new, different pollinators may be more effective, leading to a ‘pollinator shift.’ Such pollinator shifts can play a crucial role in creating reproductive isolation between the new species and its parent (Schemske & Bradshaw, 1999; Kay & Schemske, 2003; Kay et al., 2005; Kay & Sargent, 2009). Fairly simple changes in the characteristics of a population’s flowers enable them to attract bees instead of hummingbirds as pollinators, or vice versa (Schemske & Bradshaw, 1999; Kay et al., 2005).

As in any functional marketplace, rich rewards attract high-performance pollinators (Howe & Westley, 1988: 115). In Madagascar, the orchid Angraecum sesquipedale houses its rich nectar in a 30-cm spur where only the long proboscis of the far-flying sphinx moth Xanthopan morgani praedicta can reach it (Darwin, 1877: 162–166; Nilsson, 1988). Relax the intensity of competition of plants for pollinators, and the advantage of high-performance pollination vanishes. In Réunion, an island 800 km east of Madagascar with 4% of Madagascar’s area, competition with other plants, and with herbivores, is far less intense (Leigh et al., 2009). There, the related orchid Angraecum cadetii, descended from invaders from Madagascar, has a simpler flower, offers a poorer reward and attracts a flightless cricket as its only pollinator (Micheneau et al., 2010).

How does a plant persuade pollinators to carry its pollen to the next plant of its kind? Some herbs persuade pollinators to move on by offering enough of a desirable reward to attract, but not to satisfy, a pollinator, so that it looks to other plants of the same kind for more. Pollinators of such plants often ‘trapline’: they repeatedly visit, in the same order, plants whose nectar they found attractive. Here, good service attracts loyal customers who keep coming back. Other plants expel pollinators before they have eaten their fill, as male cycads expel pollinating thrips by heating up, and producing oppressive scent levels, so that they leave and carry pollen to female cycads, whose odour is now more attractive (Raguso, 2008a,b: 554). Trees with crowns full of flowers, however, would seem to face particular difficulties in causing pollinators to move on: why leave when abundant food is right at hand? Perhaps such trees attract so many pollinators that some move on to look for other trees where competition is less intense. In fact, most tropical flowering trees are self-incompatible. Most cross-pollinated species, even rare ones, maintain both their numbers and considerable genetic diversity. In many species, pollen sometimes moves hundreds of metres to fertilize conspecifics (Hamrick & Murawski, 1990, 1991; Murawski & Hamrick, 1991; Hamrick et al., 1992). How trees persuade animals to carry their pollen to other trees is worth learning.

Many plants cheat their pollinators. Trees with crowns full of bright flowers cannot cheat: it is too easy for pollinators to learn to avoid them. The story is different for rare herbs. One-third of the world’s orchid species cheat their pollinators (Cozzolino & Widmer, 2005; Schiestl, 2005). Most deceptive orchids, like deceptive plants of other species, feign a nonexistent food reward (Schiestl, 2005). In some plants, flowers mimic the shape, colour and sometimes the odour of richly rewarding flowers of other species, a ‘Batesian mimicry’ among plants (Nilsson, 1983; Johnson, 1994, 2000; Roy & Widmer, 1999). This Batesian mimicry has a twist, however: it involves superstimulus. In tropical Begonia, rewardless female flowers are larger mimics of rewarding male flowers. Experiments show that larger female flowers attract more pollinators than flowers that resemble their male counterparts more precisely (Schemske & Ågren, 1995). Other deceptive flowers bloom in places and times where masses of similarly coloured but rewarding species of plants are flowering (Johnson et al., 2003; Schiestl, 2005). Some orchids operate ‘sensory traps.’ The most usual trap attracts male insects of a certain species by mimicking, or synthesizing, the pheromones of sexually receptive females. Such orchids proffer a flower part enough like a female insect’s abdomen to excite copulation attempts by the sexually aroused male, leading to imposition of pollen on the insect, and pollination, if the insect has already tried to copulate with another orchid of this species (Schiestl et al., 1999, 2003). Here too, an exaggerated signal attracts more pollinators (Schiestl, 2005). Another orchid attracts honeybee-eating hornets by mimicking the pheromones of alarmed honeybees, causing the hornets to pounce on the flower’s red centre, which offers no food fit for it (Brodmann et al., 2009). Sensory-trap orchids attract species-specific pollinators, a circumstance that vastly accelerates their rate of speciation (Cozzolino & Widmer, 2005).

The only defence pollinators have against deceptive flowers is to avoid their like (Schiestl, 2005), or, if they cannot distinguish the deceivers from rewarding plants or real females, to move to places where deceivers are absent (Wong & Schiestl, 2002). If nonpollinating females remain unmated thanks to the repellent effects on their males of attempted copulations with orchids, they leave patches of deceptive orchids (Schiestl, 2005). This habit of ‘once bitten, twice shy’ keeps deceptive species from becoming too common (Schiestl, 2005). Indeed, orchids offering nectar rewards often evolve from food-deceptive ancestors, as if deception does not always pay. Sexually deceptive species present more complex problems. In the inflorescences of the deceptive orchid Ophrys sphegodes, flowers on different plants differ enough in odour that pollinators repelled by attempted copulation with a flower on one plant will still be attracted to another plant’s flowers (Ayasse et al., 2000). As a result of their pollinators’ attempts to avoid deceit, sensory-trap orchids outcross more often, and move their pollen to more distant counterparts, than do rewarding orchids, even though fewer of their flowers are pollinated (Cozzolino & Widmer, 2005; Schiestl, 2005).

Plants can adjust the shape, colour, size or scent of their flowers to attract desirable pollinators and repel other animals (Raguso, 2008b). This power, however, can be very limited. Many animals ‘rob’ a flower’s nectar without pollinating it. Bees and hummingbirds bite or poke holes in the bases of the tubular pink flowers of Panama’s understory treelet Quassia amara to rob their nectar. Weaker robbers use these holes to rob more nectar. These activities can reduce a Quassia’s seed production by over 75% (Roubik et al., 1985). Changes in flower characteristics, including cues that call pollinators, can sometimes reduce cheating, and attract more loyal pollinators. Like the responses to Batesian mimicry, however, these responses take place on an evolutionary time-scale. Nonetheless, plant-pollinator mutualisms drive the luxuriance, productivity and diversity of tropical forest (Corner, 1964; Regal, 1977).

Plants and their seed-dispersers

Most plants benefit if their seeds are dispersed far from their parents or other plants of their species, and the specialist pests that live on them (Janzen, 1970; Wills et al., 2006; Comita et al., 2010). Others benefit by being dispersed to light gaps or other suitable habitats (Howe & Smallwood, 1982; Howe & Westley, 1988). How do plants persuade animals to disperse their seeds? What problems do plants face in trying to do so?

Seeds are nutritious: they attract seed predators (Janzen, 1980). Plants use two basic methods to transform seed predators into seed-dispersers. One way is making seeds which animals bury for future use, in hopes that enough buried seeds will escape being eaten to replenish their species. This mutualism, originally accidental, may be essential to maintaining tree diversity in some Neotropical forests. On Barro Colorado Island, in central Panama, agoutis store long-lasting seeds against times of shortage (Smythe, 1978). Buried seeds of the black palm Astrocaryum standleyanum may rest dormant a year or two. Enough of the seeds buried by agoutis escape being eaten to maintain the black palm population (Smythe, 1989). On this island, seeds of several other species must be buried to escape attack by insects or vertebrates (Smythe, 1989; Forget & Milleron, 1991; Forget, 1993). Agoutis eat seeds or seedlings of nearly all these species. The other way to transform seed predators into seed-dispersers is to embed distasteful or poisonous seeds in fleshy fruits, like those of avocados, figs and Piper (Wheelwright, 1985a; Howe, 1986) or wrap them in tasty arils, as do nutmegs or durians (Corner, 1954; Howe & Vande Kerckhove, 1980). Both fruits and arillate seeds are intended to be eaten whole by animals that will defecate, or spit out, the seeds somewhere else. The seed fern Pachytesta illinoensis wrapped seeds in seemingly edible pulp 300 million years ago, presumably so reptiles would disperse them (Retallack & Dilcher, 1988: 1027). Some large arillate seeds, such as Virola (Forget & Milleron, 1991), and seeds embedded in a layer of sweet pulp, like Dipteryx (Forget, 1993), must be buried by agoutis even after more mobile animals disperse them.

The size, shape, texture, colour, smell, taste and placement of a fruit and its seeds influence what animals eat the fruit and disperse its seeds (Janson, 1983; Gautier-Hion et al., 1985; Howe, 1986). Fruits differ greatly in the variety of dispersers they attract (Charles-Dominique et al., 1981; Wheelwright et al., 1984). Durians and figs attract a host of species (Corner, 1954: 37: McClure, 1966); some Asian mistletoe fruits attract one or two species of flower-peckers who, in the process of defecating the seeds, glue them to a branch (Ridley, 1930: 466f). Larger fruits attract only larger animals, which disperse seeds further (Wheelwright, 1985b; Howe, 1986; Fragoso et al., 2003; Seidler & Plotkin, 2006; Blake et al., 2009). Of 37 tree species in the Côte d’Ivoire whose seeds are known to be dispersed by elephants, truly long-distance dispersers, 30 have no other disperser (Alexandre, 1978). In central Africa, elephants are keystone animals for long-distance seed dispersal of many plants (Blake et al., 2009).

How do trees persuade fruit-eaters to move seeds away from parents and their pests? We all know that eating too many strawberries becomes unpleasant. It is wise to switch to another food before this happens. Does surfeit of one kind of fruit drive fruit-eaters to trees of other kinds? In a forest of French Guiana, fruits are either sugary or fatty. Many animals need both sugar and fat, and shuttle between trees with fatty, and trees with sugary, fruits to balance their diets (Charles-Dominique et al., 1981: 405), dispersing at least some seeds away from their parents in the process. Seed dispersal is also assisted by animals that eat fruit-eaters. Boas, four-eyed opossums (Philander) and other such predators lurk in the crowns of Neotropical fruiting trees (Howe, 1979). Accordingly, bats, and even birds as big as toucans, grab a fruit or two and fly away to some safe place to eat them in peace, and drop the seeds, before returning for more (Morrison, 1980; Howe & Vande Kerckhove, 1981). Despite the circumstances favouring seed dispersal, many larger animals drop the seeds of trees they visit right under parent crowns (Charles-Dominique et al., 1981: 392; Howe & Vande Kerckhove, 1981).

Seed dispersal by animals is not always mutualistic. Most seeds attracting ants as dispersers have fat-rich appendages, elaiosomes, with a very specific odour. Some plant species cheat their ant dispersers by attaching fake elaiosomes to their seeds, which have the ant-attracting odour but no fat reward. A few tropical legumes cheat bird dispersers with seeds whose coloration mimics the rewarding aril of related legumes. Nonetheless, far fewer plants cheat dispersers than cheat pollinators (Pfeiffer et al., 2010). On the other hand, the features of fruit adapted to promote seed dispersal do not prevent many kinds of animal, ranging from weevils to parrots and white-lipped peccaries from eating seeds without dispersing any (Janzen, 1980; Howe, 1986). Such intense seed predation does not prevent trees of most species in the moist and wet tropics from bribing animals to disperse their seeds. In an everwet forest of lowland Ecuador, where competition was intense, fruits of over 90% of the tree species present were adapted to bribe animals to disperse their seeds (Gentry, 1982: 38). Here again, mutualism survives despite being much parasitized.

Evolving mutualistic ecosystems

Ecosystems are both arenas of competition and webs of interdependence. Interdependence arises when fully exploiting the resources supplied by one guild (for example, the dead vegetable matter that plants cannot help supplying) inevitably supplies resources that guild needs (such as the mineral nutrients plants need to grow). Such diffuse, accidental, costless yet often obligate, by-product mutualisms underlie ecosystem function. As life evolved over the last few billion years, innovations that triggered the spread of their inventors enhanced their ecosystem’s productivity and diversity (Fischer, 1984). Some innovators exploited new sources of energy or nutrients, some exploited old ones more effectively, some put other organisms’ wastes to productive use, some enhanced the turnover of organic matter through herbivory or predation, some enabled colonization of new habitats, some found more productive ways to deal with herbivore pressure. Each of these innovations—photosynthesis, nitrogen fixation, denitrification (which keeps the oceans from becoming nitrate brines), sulphate reduction (which keeps the land from acquiring a skin of gypsum), decomposition, especially of dead vegetable matter, predation, herbivory and the like have all provided livings for new kinds of organisms, enhancing diversity as well as productivity (Fischer, 1984; Leigh & Vermeij, 2002; Vermeij, 2004; Leigh et al., 2009). Many of these developments enhanced interdependence as well as productivity. Animals need plants, both need decomposers, all need the bacteria that maintain the world ecosystem’s chemical balance. All the world’s pathogens, many and grievous as they are, have failed to prevent increased interdependence from enhancing productivity and intensity of competition among the world’s macroscopic organisms (Leigh et al., 2009).

As evolution progressed and competition became more intense, mutualisms whose participants pooled complementary abilities to compete more effectively with others became more prevalent (Leigh et al., 2009). Although the ecosystem’s good played no role in selection for these mutualisms, both long-term and brief-exchange mutualisms have enhanced ecosystem productivity and diversity. Incorporating mitochondria allowed eukaryotes to derive more energy from their food. Incorporating cyanobacteria allowed some eukaryotes to become primary producers (Margulis, 1993). Employing mycorrhizae allowed land plants better access to soil phosphorus: indeed, mycorrhizae may have served the original land plants as their only roots (Smith & Read, 1997). Housing zooxanthellae to produce carbohydrates enabled corals to form immensely productive reefs in nutrient-poor habitats (Talent, 1988; Wood, 1993). Employing microbes as gut symbionts and culturing fungi allowed wood roaches, termites and various beetles to greatly speed wood decomposition, and thereby the recycling of carbon and nutrients (Farrell et al., 2001; Aanen et al., 2002; Inward et al., 2007; Tokuda & Watanabe, 2007).

A central problem is, do ecosystem productivity and diversity depend on mutualism? The analogy between ecosystems and human economies may point us to the answer. In human economies, some individuals cooperate with others to compete more effectively with third parties. Human societies have long used social sanctions of cheaters to restrict ‘unfair’ competition that destroys incentive for cooperation (Smith, 1759; Boehm, 1993; Sen, 2009). In larger societies, public enforcement of contracts freely arrived at helps to control cheating within cooperative undertakings (Fisher, 1930). The wider this enforcement’s reach, the more elaborate are the forms of cooperation, and the higher the economy’s productivity and diversity of occupations. Preventing cheating on freely formed mutualisms is thus perceived, surely correctly, as serving the society’s common good and enhancing its productivity. In both human societies and natural ecosystems, mutualisms are formed to exploit new opportunities or compete better with third parties, so mutualism should be as crucial to the productivity and diversity of natural as of human economies.

After the Jurassic, mutualism transformed the face of terrestrial vegetation. One ecosystem-transforming mutualism was the employment of animals as pollinators by flowering weeds, and as seed-dispersers by their descendants, which favoured fast growth over long life and costly anti-herbivore defence. Fast growth and cheap anti-herbivore defence allowed these weeds and their descendents to replace their competitors and, eventually, to dominate most forests (Regal, 1977; Leigh, 1999). As a by-product, the fast growth and cheap defence of flowering plants increased forest productivity, resource turnover, and the diversity of forest plants and animals (Corner, 1964; Sussman, 1991; Labandeira, 2006b; McKenna & Farrell, 2006). Despite the many plants and animals that parasitize these mutualisms, the unprecedented levels of cooperation among plants and animals in Cenozoic forests made these forests far more hospitable to animals. Nevertheless, flowering trees and their herbivores remain antagonists: selection favours trees that reduce herbivory without increasing costs of defence.

The newest major natural terrestrial ecosystem, grassland, evolved where vertebrate herbivores excluded trees, allowing grasses to supply them with far more abundant and accessible food, and creating a partnership of mutual benefit between grasslands and grazers (McNaughton, 1985; Retallack, 2001a: 418; West et al., 2007: 418), that might be called a diffusely coevolved mutualism. This partnership’s evolution was not driven by the ecosystem’s good. Indeed, grasses long preceded grassland (Prasad et al., 2005). Like early flowering plants, early grasses evolved a new way to cope with herbivores, only grasses perfected defence, not escape. Grass leaves grew from ground-level tips of shoots from underground rhizomes and were gritty with silica, making them hard for browsers to eat (McNaughton, 1979: 47; Retallack, 2001a: 418; 2009: 101, Fig. 10). Early grasses grew in open woodlands and scrublands (Dugas & Retallack, 1993). Large, cud-chewing mammals that could knock over trees and chew their leaves thoroughly allowed the first grasslands to evolve over thirty million years ago in places with annual rainfall between 200 and 400 mm (Retallack, 2001a: 418; 2007: 283).

In early grasslands, grass grew in discrete bunches or tussocks with no continuous root mat or sod. In Oregon, earthworms, absent from the dry sage scrublands preceding these grasslands, lived in soils of such ‘bunch grasslands’ (Retallack, 2004: 225). Mammals eating these grasses were not specialized grazers: they were better suited to eat tree and shrub leaves (Retallack, 2007: 284; 2009: 100). By nearly eliminating tree cover, these herbivores dissolved the tragedy of the commons (Falster & Westoby, 2003; Leigh, 2008: 124) created by trees building tall, expensive trunks to compete for a light level tree-destroying herbivores make freely available to grasses. The abundant light apparently favoured fast growth over costly anti-herbivore defence (cf Coley et al., 1985). In effect, large herbivores enhanced plant production and brought it within reach. Only specialized grazers, however, could fully exploit this new food.

About twenty million years ago, specialized grazers with high-crowned ‘hypsodont’ teeth reached North America (Retallack, 2004: 217) and short sod grassland–short-grass prairies–replaced bunch grassland (Retallack, 1997, 2001a, 2007). Here, grasses formed a carpet and their roots, a continuous mat, up to 40 cm deep, like the root mat of modern lawns that allows strips of sod to be rolled up and replanted elsewhere (Retallack, 2004: 226). Sod grassland soils abound in earthworms and/or other invertebrates that make their soils more fertile (Retallack, 1997: 380, 2001b: 74). Miocene fossil soils of Ft. Ternan, Kenya, resemble modern grassland soils from the Serengeti (Retallack et al., 1990) even though Ft. Ternan’s fossil grasses descended from grasses of poor soil in dry woodlands (Dugas & Retallack, 1993). Fire maintains some grasslands, especially those with few grazers (Dublin, 1995), but fire played little role in the evolution of short sod grasslands in Oregon and Nebraska (Retallack, 2004: 231).

In short-grass prairie, large animals cannot hide: they must run to catch prey or escape predators. To cope with pack-hunting predators, grazers formed herds. The first fossil herd of grazers appeared on newly evolved short sod grassland in Nebraska nineteen million years ago (Retallack, 2007: 285). Did predators lead a herd of grazers to graze a patch of grassland, palatable plants and unpalatable, down to the ground, and manure it thoroughly, before moving on to the next? This mode of grazing has been called the best way to raise cattle on natural grassland (Retallack, 2007: 285). Seven million years ago, tall sod-forming grassland, tall-grass prairie, which formed root-mats up to a metre deep, replaced forests in settings with annual rainfall between 400 and 750 mm (Retallack, 1997, 2001a,b, 2007). Since then, grasses have expanded into somewhat wetter climates (Retallack, 2001a).

The Serengeti is the most spectacular, least changed and most studied of the surviving grassland ecosystems (Sinclair & Norton-Griffiths, 1979; McNaughton, 1985; Sinclair & Arcese, 1995). Grasses and vertebrate herbivores were originally antagonists. As we have seen, the evolution of grasses, grazers and predators, each in its own interest, later made the grass-grazer relationship far more mutualistic (Retallack, 2001b: 316), as if an ecological counterpart of Adam Smith’s invisible hand were at work. Serengeti’s grasslands need elephants to exclude trees (Dublin, 1995), and many grazers to maintain the grassland’s productivity, diversity and soil fertility (McNaughton, 1979, 1985). Although grazers reduce production of grass seeds by 90%, they increase vegetative reproduction (McNaughton, 1979: 47; 1985: 273). The grazers are adapted in many ways to maintain high grass production. The movement patterns and feeding habits of different grazer species, especially the wall of wildebeest that follows the rains and the new grass back and forth across the Serengeti, are crucial to maintaining grass production (Maddock, 1979; McNaughton, 1985). Grazers preferentially manure favoured treeless habitats and keep them in arrested succession where they supply nutritious new growth as long as the soil stays moist (McNaughton, 1985). Grazers thus derive far more copious and nutritious food from the Serengeti than vertebrate folivores obtain from any rainforest. Serengeti grazers eat an average of 66% of its grasses’ above-ground production, whereas in a rainforest, vertebrates eat < 10% (McNaughton, 1985: 283). Grass production would be even higher with somewhat fewer grazers (McNaughton, 1985: 284): grasses load their root collars with silica to prevent overgrazing (McNaughton et al., 1984). Even though there is little sign that either grasslands or grazers evolved to benefit each other, the relationship between grasslands and grazers is somewhat like that between farmers and their crops, and more like that between certain damselfish and the algal turfs on which they feed and which they protect from grazing sea urchins (Eakin, 1987).

Concluding remarks

Mutualism arises when members of two or more species benefit from each other’s activities. Costly mutualism involves members of two or more species cooperating by behaviours selected each for the benefit of the other species. Costly mutualism evolves only if it benefits all participants. In most mutualisms, members of different species pool complementary abilities for their mutual benefit by exchanging one good or service for another. Such mutualisms fall into two categories (Table 1): symbioses, where members of at least one partner species spend most of their lives closely associated with the other, and mutualisms of brief exchange, where participants cooperate briefly and go their separate ways. Biologists concerned with how cheating is suppressed have focused largely on symbioses: as we have seen, they have learned how such symbioses are maintained. True, some symbioses, such as those between figs and their pollinating wasps (Herre et al., 2008) or between mutualistic sponges (Wulff, 2008), are parasitized, sometimes heavily, by third parties. How these parasites are limited to tolerable levels is an open question.

Most interactions of plants with pollinators or seed-dispersers are mutualisms of brief exchange that increase ecosystem productivity, diversity and hospitability to animals (Regal, 1977). We know far less about how cheating on and parasitism of these mutualisms is restrained enough that they remain advantageous. Many plants may only be able to discourage cheating by designing the structure and appearance of flowers or fruit and the nature or placement of their rewards to more nearly attract only those animals most likely to perform the needed service. Yet understanding how mutualisms of brief exchange survive the attentions of cheaters and parasites may be crucial to understanding how ecosystems became more mutualistic.

Indeed, mutualisms are essential to the world ecosystems’ productivity and diversity. Cooperative enterprises are so obviously vital to human economies that laws are enforced to prevent cheating within them. Similarly, mutualisms must benefit natural ecosystems. In tropical forest, flowering plants outgrew competitors by forming mutualisms with animals that reduced the plants’ defence costs. Forest productivity and diversity increased as an incidental result. In drier climates, novel, ferocious defences of grasses against forest-destroying browsers led to the evolution of grassland where grazers benefited from a vastly greater supply of nutritious food and grasses from grazers excluding competing trees. Here, folivores and plants have become allies, foreshadowing, however accidentally, the relationship between farmers and the plants they cultivate. Is the increase of ecosystem productivity through multiplication of mutualisms a general trend?


I am most grateful to Angela Douglas, Margaret McFall-Ngai and Janie Wulff for helping me understand the many ways symbiotic mutualisms can be enforced and to Anthony Coates and A. G. Fischer for helping me understand the role of photosymbiotic symbionts in the evolutionary history of coral reefs. G. J. Vermeij and E. Allen Herre provided useful criticisms of the manuscript and a host of useful references. Vielka Chang-Yau and Angel Aguirre, librarians of the Smithsonian Tropical Research Institute, procured the internet-illiterate author a host of pdfs crucial for writing this paper. Both the editor and this paper’s reviewers helped me understand what I was trying to say, which improved the paper greatly. I am most indebted to G. F. Handel for providing an atmosphere that inspires writing and to the plants and animals of Barro Colorado Island for their continual reminder of what I should try to understand.