Skip to content
NatureHQ

Ecologyecological relationship

Mutualism

A cleaner fish would rather eat its client’s mucus than its parasites. Mutualism is not friendship — it is an arrangement that needs holding in place.

Two species each getting something out of the other. Not friendship: the partners want different terms, cheating is common, and the interesting question is not why they cooperate but what stops the cheating from wrecking it.

The word invites the wrong picture. Mutualism does not mean two species helping each other; it means each doing something for its own reasons that happens to benefit the other, and the interests behind those reasons are rarely aligned for long. The cleaner wrasse states the problem in one sentence: its preferred food is not the parasites on its client but the client’s own mucus, which the client does not want removed. So the honest behaviour needs an explanation, and part of the answer is surveillance — cleaners take parasites rather than mucus more often when another client is watching, and clients prefer cleaners they have seen behaving well. Cheating in front of the queue costs future custom. Read that way, mutualisms look less like partnerships and more like arrangements held in place by something: sanctions, partner choice, or the difficulty of defecting unnoticed. Exploitation turns up in nearly every mutualism examined closely, and the interactions persist anyway — which is the thing needing explanation. The other half of this page is a claim that survived. Honeyguides, according to a story told for centuries, lead people to bees’ nests so that the people open the hive and the bird gets the wax. It sounds exactly like the sort of account this site usually has to correct. It is true. Guided searches find nests dramatically faster than unguided ones, and the birds’ flight direction and perch height track where the nest is. And the modern work went further than the folk version: honey-gatherers use a specialised call to recruit birds, and playing it back more than doubles the chance of being guided, against controls including other human speech. The signalling runs both ways — and the bird learned what a human noise means despite being a brood parasite, raised by strangers.

Early coverage · 63% complete · reviewed 2026-09-03

What this page covers

Mutualisms occur across all of life and many of the most consequential cross kingdoms: plants with fungi, animals with bacteria, plants with pollinators. The cases here are animal ones because they are where the experiments are cleanest.

Often confused with: Symbiosis, which means living together and covers parasitism too; Cooperation, which implies a shared interest that mutualism does not require; A stable arrangement being a harmonious one, when stability is what needs explaining

Quick facts

Not cooperation
Conflicting interests, and cheats in nearly every system examined
Held in place by an audience
Cleaners behave better when another client is watching
And it does work
Remove the cleaners and parasite loads rise
A folk claim that held
Honeyguides really do lead people to bees’ nests

Not friendship, an arrangement

The partners want different terms, and the cheating needs explaining away.

Mutualism is not harmony. Each partner is in it for itself, the two want different terms, and cheats — partners and outsiders — turn up in nearly every mutualism anybody has examined closely.

Established

Specialists would state this without hedging. Multiple independent lines of evidence agree.

Documented mutualisms generally contain exploitation, by partners taking benefit without paying cost and by third parties extracting benefit from outside the interaction. Persistence is attributed to mechanisms including partner sanctions, partner choice and spatial structure rather than to an absence of conflict.

Who this applies to
Mutualistic interactions generally, across kingdoms.
Studied in
Animalia, Plantae, Fungi
Why we rate it this way, and what the caveats are
EstablishedHigh confidence

A settled position in the ecological literature, supported by many independent systems.

How far it can be extended

Exploitation is documented across most well-studied mutualisms.

Caveats

  • Saying partners have conflicting interests does not mean the interaction is unstable — most are stable, which is the thing needing explanation.
  • Study effort concentrates on mutualisms that are easy to manipulate, so exploitation rates may not be representative.

Still unanswered

  • Why some mutualisms tolerate high levels of cheating and others appear not to.

Last reviewed 2026-09-03

The evidence (2 studies)

A cleaner wrasse would rather eat its client’s mucus than its parasites. It takes the parasites more often when another client is watching — and clients prefer cleaners they have seen behaving well.

Well supported

Good evidence backs this, though some details remain open.

Cleaner wrasse increased cooperative feeding — taking ectoparasites rather than client mucus — in the presence of an observing client, and clients preferentially approached cleaners observed behaving cooperatively.

Who this applies to
Demonstrated under aquarium conditions with controlled observation.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Labroides dimidiatus
Why we rate it this way, and what the caveats are
Well supportedModerate confidence

A controlled experiment with a clear effect. How much of natural cleaning behaviour it accounts for is a separate question.

How far it can be extended

Aquarium conditions differ from a reef in how unavoidable an audience is.

Caveats

  • An aquarium audience is present in a way a reef audience may not be.
  • This shows an effect of observation, not that reef cleaners are generally honest.

Still unanswered

  • What a client actually tracks — an individual cleaner, or a station.

Last reviewed 2026-09-03

The evidence (1 study)

The cleaner-fish result is worth sitting with. The cleaner’s preferred food is the thing its client is least willing to give up, so every interaction contains a small conflict, and honest behaviour is not the default that occasionally lapses — it is the outcome that has to be produced. What produces it here is other fish waiting their turn. A cleaner that cheats in front of the queue loses custom, and the clients do watch.

Living off another organism, as this corpus divides it
RelationshipWhat the other organism gets
PredationKilled and eaten
ParasitismKept alive and drained
ParasitoidismConsumed entirely, slowly
Brood parasitismA season of work, spent on someone else’s young
MutualismSomething it wanted — on terms it would rather improve

The rest of the set

Somebody finally removed the cleaners

An obvious claim that had gone untested for decades.

Remove the cleaner wrasse from a patch of reef and the fish there end up carrying more parasites. The obvious claim had gone untested for decades before somebody removed them and counted.

Well supported

Good evidence backs this, though some details remain open.

Experimental removal of cleaner wrasse from reef patches resulted in substantially higher parasite loads on client fish relative to control patches where cleaners remained.

Who this applies to
Particular reef patches and client species; effect sizes differ between systems.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Labroides dimidiatus
Why we rate it this way, and what the caveats are
Well supportedModerate confidence

A clean field manipulation with controls. Removal also changes client movement, so not every difference is cleaning alone.

How far it can be extended

Later studies report differing effects depending on reef, client species and duration.

Caveats

  • Removing cleaners changes more than cleaning: which fish visit the patch, and how often, changes too.
  • Effect sizes vary between later studies, reefs and client species.

Still unanswered

  • How much of a client’s parasite burden cleaning removes over a lifetime, as against a season.

Last reviewed 2026-09-03

The evidence (1 study)

How we know

Taking the cleaners off a reef

Cleaner fish are seen picking parasites off other fish. Does that measurably reduce how many parasites those fish carry?

Cleaner wrasse were removed from experimental reef patches and kept absent for an extended period, with comparable patches left intact as controls. Parasite loads on client fish were then measured on both sets of patches.

What happened

Client fish on the cleaner-free patches carried substantially more parasites than those on control patches.

What it shows

That the interaction does what it appears to do. That may sound unremarkable, and the reason it is not is that everybody had assumed it for decades without testing it: fish visiting cleaners, cleaners picking, and parasites on the reef were all documented, and the connection between them was not.

What it does not show

Removing cleaners changes more than cleaning — which fish visit the patch and how often changes too, so not every difference is attributable to parasite removal alone. Effect sizes also vary between reefs, client species and later studies.

The controls — what makes this evidence rather than a story
  • Matched control patches with cleaners left in place, so the comparison is cleaning rather than reef.
  • A removal long enough for parasite loads to respond, rather than a snapshot.
  • Parasites counted directly on client fish rather than inferred from behaviour.

From Cleaner fish really do clean

The pattern is one this site keeps finding. Everything visible was documented — fish queueing at cleaning stations, cleaners picking, parasites present on the reef — and the connection between them was assumed rather than measured. It took removing the cleaners from patches of reef and counting parasites afterwards to establish that the interaction has the effect everybody had been describing.

What happens when the reason for the arrangement goes away

An acacia, its guard ants, and a fence.

Fence the large herbivores out of a savanna and, within a decade, acacias stop paying their guard ants. A non-defending ant species takes over, and the trees grow more slowly and die more often than the browsed ones.

Well supported

Good evidence backs this, though some details remain open.

Decade-long experimental exclusion of large herbivores from African savanna plots reduced acacia investment in ant rewards, shifted ant occupancy from a defending species to a less protective one associated with wood-boring beetles, and resulted in reduced tree growth and increased mortality relative to unfenced controls.

Who this applies to
One African savanna system and its particular acacia–ant associations.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Acacia drepanolobium, Crematogaster
Why we rate it this way, and what the caveats are
Well supportedHigh confidence

A decade-long manipulation with unfenced controls, in which the mechanism — reduced rewards, changed occupancy, worse tree outcomes — was measured at each step.

How far it can be extended

A specific set of partners in one system; other ant-plant mutualisms differ in structure and rewards.

Caveats

  • Exclusion removes several herbivore species at once and changes more than browsing pressure.
  • One system; the result shows that a mutualism can degrade this way, not that all will.

Still unanswered

  • Whether the degradation reverses if herbivores return, and over what timescale.

Last reviewed 2026-09-04

The evidence (2 studies)

How we know

Fencing out the browsers, and watching a partnership fail

Acacias feed and house ants that defend them from browsers. What happens to the arrangement if the browsers are removed?

Large herbivores were excluded from plots of African savanna for a decade, with unfenced plots as controls. Within both, acacia investment in ant rewards was measured — swollen thorns and nectar — along with which ant species occupied each tree, and the trees’ growth and survival.

What happened

Trees behind the fences produced fewer swollen thorns and less nectar. Occupancy shifted from the defending ant species to one that tolerates lower rewards, does not defend the tree, and is associated with wood-boring beetles. Those trees grew more slowly and died more often than the browsed ones outside.

What it shows

That a mutualism is held in place by the conditions that make it worth paying for. Remove the threat and the payments fall; lower payments admit a partner that provides nothing; and the tree ends up worse off than when it was being eaten. It is the clearest demonstration in this corpus that partnership outcomes are conditional rather than fixed properties of the species involved.

What it does not show

Exclusion removes several herbivore species at once and changes more than browsing — light, grass competition and soil all shift behind a fence. It is one savanna, one acacia and its particular ants, and it shows that a mutualism can degrade this way rather than that others will. Whether the arrangement recovers if browsers return is not established.

The controls — what makes this evidence rather than a story
  • Unfenced control plots in the same savanna, so the comparison is browsing pressure rather than site.
  • A decade of exclusion, since the tree’s investment and the ant community change over years rather than seasons.
  • Rewards, occupancy and tree outcome all measured, so the chain is followed rather than inferred from its ends.

From Breakdown of an ant-plant mutualism follows the loss of large herbivores

The acacia in this system houses ants in swollen thorns and feeds them nectar, and the ants attack anything that starts eating the tree — which, in an African savanna, means elephants and giraffes. The arrangement is expensive for the tree, and it is worth the expense because something is trying to eat it.

Fence the browsers out for a decade and the tree reduces its payments. Lower rewards let a different ant species take over — one that tolerates poor provisioning, does not defend the tree, and is associated with wood-boring beetles. Trees inside the fences grew more slowly and died more often than the trees still being browsed outside. This is the strongest experimental evidence in this corpus that a mutualism is conditional: nothing about either partner changed, and removing the threat the partnership answered was enough to degrade it.

The same lesson elsewhere

A folk claim that survived testing

And then turned out to be more than the folk version said.

A wild bird leads human honey-gatherers to bees’ nests, and the guiding measurably works: guided searches find nests far faster than unguided ones, and the bird’s behaviour tracks where the nest is.

Established

Specialists would state this without hedging. Multiple independent lines of evidence agree.

Greater honeyguides guide human honey-gatherers to bees’ nests. Guided searches locate nests substantially faster than unguided searches, and the birds’ flight direction and perch height correspond to the direction and distance of the nest.

Who this applies to
Documented with particular honey-gathering communities in East and southern Africa.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Indicator indicator
Why we rate it this way, and what the caveats are
EstablishedHigh confidence

Quantified against unguided searches, with the birds’ behaviour related to nest position, and later confirmed experimentally.

How far it can be extended

The interaction depends on local human practice as much as on the bird, and both differ between regions.

Caveats

  • Whether the bird intends to inform is not established, and behaviour alone cannot settle it.
  • The practice is declining with changes in honey-gathering, so the interaction is not uniform across its range.

Still unanswered

  • Whether honeyguides guided any other animal before people, which is proposed and unresolved.

Last reviewed 2026-09-03

The evidence (2 studies)

Honey-gatherers use a specialised call to recruit the birds — and playing it back more than doubled the chance of being guided, against other human sounds. The signalling runs both ways.

Well supported

Good evidence backs this, though some details remain open.

Playback of the specialised recruitment call used by Yao honey-gatherers more than doubled the probability of being guided by a greater honeyguide relative to matched control sounds, including other human vocalisations, and increased the probability of locating a bees’ nest.

Who this applies to
One human community and one region; recruitment calls differ between honey-gathering cultures.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Indicator indicator
Why we rate it this way, and what the caveats are
Well supportedHigh confidence

A field playback experiment with the right controls — other human sounds, which separate the specific call from human presence.

How far it can be extended

The call is culturally specific, and whether birds elsewhere respond to different calls is a separate question.

Caveats

  • It establishes that birds respond to the call, not how they come to learn it.
  • One community and one call; the generality across honey-gathering cultures is untested.

Still unanswered

  • How young honeyguides learn which human sound is worth answering, given they are raised by other species.

Last reviewed 2026-09-03

The evidence (1 study)

How we know

Playing a human call to a wild bird

Honey-gatherers use a particular call to summon honeyguides. Do the birds respond to that call, or simply to people being in the bush?

The specialised recruitment call used by Yao honey-gatherers was played back along transects in honeyguide habitat, alongside control sounds matched for the situation — other human vocalisations, including words shouted in the same voice, and the call of a different bird species. For each trial, whether a honeyguide arrived, stayed and led the party to a bees’ nest was recorded.

What happened

The specialised call more than doubled the probability of being guided compared with the controls, and raised the probability that a trip ended at a bees’ nest.

What it shows

That the signalling runs in both directions. The bird signals to the human, which was already documented, and the human signals to the bird, which the bird has learned to answer. It is among very few documented cases of a wild animal responding to a human signal to mutual benefit, without training and without captivity.

What it does not show

It does not show how a young honeyguide learns which sound is worth answering — a real puzzle, since honeyguides are brood parasites raised by other species and never meet a knowledgeable parent. One community and one call; whether birds elsewhere answer different calls is untested.

The controls — what makes this evidence rather than a story
  • Other human sounds, which separate the specific call from human presence — the alternative that would otherwise explain everything.
  • Another bird’s call, controlling for a response to any biological sound.
  • The same walkers, route type and conditions across treatments, so the sound is the variable.

From Reciprocal signaling in honeyguide-human mutualism

It is worth being explicit about why this is unusual. A great deal of what this site does is establish that a vivid, widely repeated claim about animals does not survive measurement. Here is one that did — and the measurement then found something the story had not: the birds respond to a specific human call, distinguishing it from other human sounds. The traditional account had the bird recruiting the person. It runs both ways.

The research behind this page

9 studies, newest first. Each one has a page explaining what it found and what it could not show.

2016Science

Reciprocal signaling in honeyguide-human mutualism

The specialised call more than doubled the probability of being guided compared with control sounds, and increased the probability of the trip ending at a bees’ nest.

2014Science

Niche engineering demonstrates a latent capacity for fungal-algal mutualism

Under those conditions the two organisms grew together where neither grew alone, exchanging carbon dioxide and nitrogenous compounds.

2008Science

Breakdown of an ant-plant mutualism follows the loss of large herbivores

With large herbivores excluded, trees reduced their production of swollen thorns and nectar.

2006Nature

Image scoring and cooperation in a cleaner fish mutualism

Cleaners behaved more cooperatively — taking parasites rather than mucus — when an observing client was present, and clients preferred to approach cleaners they had observed behaving cooperatively.

2002Animal Behaviour

Asymmetric cheating opportunities and partner control in a cleaner fish mutualism

Clients responded to being bitten by leaving or by chasing the cleaner.

2001Ecology Letters

The exploitation of mutualisms

Exploitation is present in most well-studied mutualisms, by partners and by outsiders, and mutualisms nevertheless persist.

1999Nature

Cleaner fish really do clean

Client fish on reefs without cleaners carried substantially more parasites than those on control reefs.

1989Science

Honeyguides and honey gatherers: interspecific communication in a symbiotic relationship

Searches accompanied by a honeyguide located nests substantially faster than searches without one, and features of the bird’s behaviour — including flight direction and perch height — corresponded to direction and distance to the nest.

1879Karl J. Trübner

Die Erscheinung der Symbiose

Associations between unlike organisms form a continuum of outcomes, and the fact of living together does not determine which outcome applies.

This page is a stop on a longer route

A guided journey reads several subjects in a deliberate order, with an argument for why one follows another. You can join in the middle.

Where to go from here

Each of these follows from something on this page — a relationship in the evidence, a claim people ask about, or the next mechanism along.

How complete this page is, and what it is still missing

NatureHQ publishes its own gaps. This page is at 63% completeness against what we would call a finished subject, and was last reviewed on 2026-09-03. It carries 9 claims and answers 0 mapped search questions.

  • no research from the last few years is attached — check for newer work
  • no popular claim about this subject has been checked yet
  • Mycorrhizal and gut-microbial mutualisms are far more consequential than the animal cases here and are covered elsewhere or not at all.
  • Ant–plant and ant–aphid mutualisms, which are among the best-studied, are not treated.
  • The evolutionary origins of mutualisms — how they start — is not covered.