Mutualism, Parasitism and Coevolution

A fungus that helps a plant grow in poor soil can slow the same plant down in rich soil. So is the fungus a mutualist or a parasite? This lesson treats that as a question you answer with data, and then asks a harder one: when two species interact for many generations, how can you tell whether each has evolved in response to the other?

The idea: interaction signs are outcomes

Ecologists name interactions by the net effect on each partner’s fitness, its survival and reproduction relative to the alternative. In a mutualism (+,+) both partners gain. In commensalism (+,0) one gains and the other shows no net change. In parasitism (+,-) the parasite gains and the host loses. Predation, covered in predation, shares the parasite’s sign but usually kills the prey outright.

Treat each sign as the reading on a balance. A zero in commensalism is a net value: small costs and small benefits can cancel, and Mathis and Bronstein’s 2020 review argued that commensalism can persist as a stable outcome rather than a waystation to something else.

Because the sign is a net, it can move. Mycorrhizal fungi grow into plant roots, take carbon from the plant and supply soil nutrients, especially phosphorus. Johnson, Graham and Smith (1997) argued that when the plant’s costs exceed its benefits, as can happen when nutrients are plentiful, the fungus might be considered parasitic from the plant’s side. The same pair sits somewhere on a mutualism-parasitism continuum, and where depends on the soil.

Microbes show it too. Shapiro, Williams and Turner (2016) found that in early tests Escherichia coli infected with the filamentous phage M13 grew faster and reached higher density than uninfected cells, which they read as a facultative mutualism; after experimental evolution the phage no longer helped the ancestral strain. A virus is usually filed as a parasite; the measurement said otherwise there.

Mutualisms and how hosts keep partners honest

A mutualism has a weak point. If a partner can take the reward and skip the service, that partner may leave more offspring than an honest one, and cheaters could spread. Rhizobia are bacteria that live in root nodules of legumes and fix atmospheric nitrogen for the plant in exchange for carbon. What stops non-fixing strains from taking over?

One answer is host sanctions: the host reduces the fitness of partners that do not deliver. Kiers, Rousseau, West and Denison (2003) tested this in soybean, Glycine max, with Bradyrhizobium japonicum. Comparing a fixing strain with a non-fixing one would be confounded, because the strains differ at many genes. So they kept the strain and removed the service, replacing the nitrogen in the air around roots or single nodules with argon. Rhizobial reproductive success fell by about 50% in nodules that could not fix. Reduced oxygen supply to those nodules was implicated as a possible mechanism; it was not shown to be the cause. The 50% figure belongs to soybean in that study, not to every legume.

“Sanction” names a measured effect, not a decision. The plant does not judge. Plants that cut the reproduction of non-fixing nodules, by whatever mechanism, leave fewer cheaters in the soil, and that effect is what selection can maintain.

Sanctions also depend on context. Fronk and colleagues (2025) grew Lotus japonicus with fixing and non-fixing rhizobia. Under nitrogen fertilization the plants no longer discriminated between them, and they reduced their overall investment, making small nodules with fewer rhizobia. When the plant gets nitrogen from the soil, the rhizobium’s service is worth less, and the line between partner and freeloader blurs.

Parasites, hosts and two kinds of coevolution

Matching traits are tempting evidence. A long-tongued moth and a deep flower look like a matched pair. Daniel Janzen’s 1980 note “When is it coevolution?” set a stricter standard: coevolution is evolutionary change in one population in response to another, followed by an evolutionary response in the second. A good fit can arise when only one partner evolved, so the fit alone is not the evidence.

Hosts and parasites give the clearest tests, because both have short generations and strong effects on each other. Two kinds of dynamics are distinguished:

• An arms race is escalation: host resistance and parasite infectivity both rise steadily through time.
• Fluctuating selection is cycling: whichever host genotype is common is attacked most, so rare genotypes gain, and the advantage keeps shifting. This is negative frequency-dependent selection, whose arithmetic is in population genetics.

The analogy is two chess players. In an arms race both keep getting stronger, so a player from today would beat either player from last year. In fluctuating selection each player has learned only the other’s current habits, so today’s version does best against today’s opponent.

A time-shift assay puts that analogy to work. You freeze samples through time and cross hosts from one time with parasites from the past, the present and the future. Betts, Kaltz and Hochberg (2014) did this with Pseudomonas aeruginosa PAO1 and six lytic phages, and five of the six gave patterns consistent with frequency-dependent selection, while two of the six also showed directional selection. Paterson and colleagues (2010) let the phage phi2 evolve with Pseudomonas fluorescens SBW25 or with a host held at a constant genotype. Sequencing showed faster phage molecular evolution, and more divergence between replicate populations, when the host coevolved. Phage biology itself is in bacterial and phage genetics.

Natural systems can be tested across time when dormant stages are stored. Decaestecker and colleagues (2007) used dormant Daphnia eggs and parasite spores from dated layers of pond sediment, a method called resurrection ecology. Parasites had adapted to their hosts within a few years, and virulence, the harm done to an infected host, increased through the record. In that Daphnia system the result ran against the common claim that parasites evolve toward harmlessness, though it comes from one host and one parasite. Myxoma virus in Australian rabbits is the classic field case: after its release around 1950 its virulence fell to intermediate levels, not to zero.

For human health, these laboratory and field systems show mechanisms, not treatments. Pathogen variants that escape host defences, such as those in Toll-like receptor signalling and antigen receptors, can leave more descendants, which puts those defences under reciprocal selection. Experiments in the nematode Caenorhabditis elegans (Rafaluk-Mohr and colleagues, 2018) found that a resident gut bacterium and its host coevolved toward a more protective partnership when a pathogen was present. That is a worm result, not evidence for any human probiotic or phage therapy.

Ecology and evolution on one clock: rotifers and algae

In classical predator-prey models, the predator peak lags the prey peak by about a quarter of a cycle: prey rise, predators feed and multiply, prey crash, predators starve. The phase lag is the delay between peaks as a fraction of the cycle’s period. Yoshida, Jones, Ellner, Fussmann and Hairston (2003) grew the rotifer Brachionus calyciflorus on the alga Chlorella vulgaris in nitrogen-limited chemostats, with the algae as either one clone or several. One clone cannot evolve by selection among clones; several can. They reported that single-clone prey gave short cycles with the classical quarter-cycle lag, while multiclonal prey gave long cycles with predator and prey peaks largely out of step.

Worked problem. Here is an invented data set. Assume the cycles are regular, the period does not change during the run, and each predator peak is matched with the prey peak just before it. (In the real experiment the treatments did not share a period; here they do, to keep the arithmetic clean.)

• Run A: prey peak on days 10, 40 and 70; rotifers peak on days 17.5, 47.5 and 77.5.
• Run B: prey peak on the same days; rotifers peak on days 25, 55 and 85.

1. Period. 40 - 10 = 30 days, and 70 - 40 = 30 days. The period is 30 days.
2. Lag in run A. 17.5 - 10 = 7.5 days, so 7.5 / 30 = 0.25 cycle. That is the classical quarter lag.
3. Lag in run B. 25 - 10 = 15 days, so 15 / 30 = 0.50 cycle. Predator peaks sit at prey troughs: antiphase.

phase lag (cycles) = delay between peaks (days) / period (days)

Predator delay 7.5 days

Run A (delay 7.5 d) Run B (delay 15 d) Practice case (24-day cycle, delay 12 d)

Drag the markers (or focus one and press the arrow keys; Page Up and Page Down move 5 days): P1 and P2 on two successive prey peaks, R on the predator peak just after P1.

▼ model lag   ▲ your measured lag

Model (expected)

Your markers (observed)

In this sketch: regular cycles, a fixed period within each run, abundance in relative units (not to scale), and curves drawn from the lag alone. Runs A and B and the practice case are invented numbers, not Yoshida’s data; in the real experiment the two treatments did not share a period. The two cases are named in words only, because the lesson gives no numeric ranges.

Rule to keep: phase lag (cycles) = delay between peaks (days) / period (days), with each predator peak matched to the prey peak just before it. The lag alone does not prove that the prey evolved.

4. Compare. Run A’s 0.25 is the classical quarter lag that the single-clone runs showed. Run B’s 0.50 is antiphase, like the long multiclonal cycles whose peaks were largely out of step. Run B resembles the evolving-prey runs.
5. Limit. The lag alone does not prove that algae evolved. A long lag could have other causes. The causal case came from manipulating clonal diversity, and from a measured trade-off: algal clones that resisted grazing were poorer competitors for nutrients.

The feedback runs like this. Grazing favours defended clones; as they spread, food for rotifers falls; rotifers decline; with grazing relaxed, faster-growing undefended clones regain ground, and the cycle repeats. That trade-off is the same kind you met in competition, and the general case is in selection in an ecological context. One point needs care: the 2003 paper showed prey evolution and had not investigated predator evolution. This is eco-evolutionary feedback, not demonstrated coevolution.

Design problem

You have frozen samples of a bacterium and its phage taken every two weeks for 20 weeks. Design a test that separates an arms race from fluctuating selection.

One way to do it

Revive hosts and phage from all 11 time points (weeks 0, 2, 4 up to 20). For each phage time point, challenge hosts from the past, the same time and the future, for example phage from week 10 against hosts from weeks 4, 10 and 16. Measure infection as the proportion of host colonies a phage lyses, with several colonies per host-phage pairing.

Predictions. Under an arms race, phage infect past hosts best and future hosts worst, and hosts resist past phage best; the effect grows steadily with the time gap. Under fluctuating selection, infection peaks on contemporary hosts and falls off in both directions, with no long-run trend.

Controls. Run several replicate lines, since one line could follow either path by chance. Add Paterson’s constant-host treatment, phage evolving on a host held at one genotype, so you can tell reciprocal change from phage-only adaptation. Betts and colleagues (2014), using a time-shift design with Pseudomonas aeruginosa PAO1, found patterns consistent with frequency-dependent selection for five of six phages. The result applies to those laboratory populations and says nothing about phage therapy.