Widespread fitness alignment in the legume–rhizobium symbiosis


  • Maren L. Friesen

    1. Center for Population Biology, University of California, Davis, One Shields Ave., Davis, CA 95616, USA
    2. Present address: Section of Molecular and Computational Biology, Department of Biology, University of Southern California, 1050 Childs Way, RRI 201-B Los Angeles, CA 90089, USA
    Search for more papers by this author

Author for correspondence:
Maren L. Friesen
Tel: +1 213 740 3065
Email: friesen@usc.edu


  • Although ‘cheaters’ potentially destabilize the legume–rhizobium mutualism, we lack a comprehensive review of host–symbiont fitness correlations.
  • Studies measuring rhizobium relative or absolute fitness and host benefit are surveyed. Mutant studies are tallied for evidence of pleiotropy; studies of natural strains are analyzed with meta-analysis.
  • Of 80 rhizobium mutations, 19 decrease both partners’ fitness, four increase both, two increase host fitness but decrease symbiont fitness and none increase symbiont fitness at the host’s expense. The pooled correlation between rhizobium nodulation competitiveness and plant aboveground biomass is 0.65 across five experiments that compete natural strains against a reference, whereas, across 14 experiments that compete rhizobia against soil populations or each other, the pooled correlation is 0.24. Pooled correlations between aboveground biomass and nodule number and nodule biomass are 0.76 and 0.83.
  • Positive correlations between legume and rhizobium fitness imply that most ineffective rhizobia are ‘defective’ rather than ‘defectors’; this extends to natural variants, with only one significant fitness conflict. Most studies involve non-coevolved associations, indicating that fitness alignment is the default state. Rhizobium mutations that increase both host and symbiont fitness suggest that some plants maladaptively restrict symbiosis with novel strains.


Mutualisms are fundamental to ecology and agriculture, with symbiotically fixed nitrogen from the legume–rhizobium interaction providing one-third of the protein in the global human diet (Graham & Vance, 2003) and insect pollination required for three-quarters of crops (Potts et al., 2010). Global change threatens many of these fundamental services through habitat loss, the introduction of novel species and the alteration of the abiotic environments in which species interact, yet we are only beginning to understand the evolutionary forces that shape mutualisms and govern their ability to respond to a changing world (Kiers et al., 2010). The interaction between legumes, species in the family Fabaceae, and the polyphyletic group ‘rhizobia’, which includes members of alpha-proteobacteria and beta-proteobacteria (Velázquez et al., 2010), is a model system for understanding the ecological and evolutionary basis of mutualism (De Mita et al., 2007; Heath, 2008; Kiers & Denison, 2008). Multiple species’ genomes of both the legume host and the bacterial symbiont have been or are in the process of being fully sequenced (Mavingui et al., 2002; Young et al., 2006; Amadou et al., 2008; Sato et al., 2008; Schmutz et al., 2010; Branca et al., 2011; Young et al., 2011) and there are many genetic tools to dissect the molecular basis of symbiosis in these organisms (Anéet al., 2008; Young & Udvardi, 2009; Friesen & von Wettberg, 2010).

The ecological definition of mutualism requires that increasing the population size of one partner causes an increase in its partner’s population size, whereas the individual-level definition requires that interaction with a mutualist partner increases the focal individual’s relative fitness (Boucher et al., 1982). The legume-rhizobium interaction shows evidence of both rhizobium inoculation consistently having a beneficial effect on legume growth (Kaschuk et al., 2010) and the growth of a compatible host increasing the relative abundance of rhizobia in the soil (Kuykendall, 1989). However, within an interaction that is mutualistic overall, it is possible for antagonistic coevolution to occur as partners may experience conflicts of interest. The parameters of the interaction that maximize fitness for the host are not necessarily those that maximize symbiont fitness. For example, one common trait of rhizosphere-inhabiting microorganisms is the ability to synthesize the plant hormone auxin (Patten & Glick, 1996), which alters the root : shoot ratio and could deviate from the plant’s optimal phenotype (Friesen et al., 2011). In the legume-rhizobium interaction, there may be fitness conflict over the exchange of resources if individuals can ‘cheat’ by withholding the resource they produce (rhizobium-fixed nitrogen or plant-fixed carbon) whilst continuing to accept resource inputs by their partner.

There are several lines of evidence that support the plausibility of ‘defector’ rhizobia, namely, strains with high fitness that provide little or no host benefit. First, naturally occurring rhizobium strains vary in how much nitrogen they fix on a given host genotype; this benefit can differ across multiple host genotypes or species (Mytton, 1975; Burdon et al., 1999; Thrall et al., 2000, 2007; Heath & Tiffin, 2007). Second, over half a century of genetic work has identified many genes involved in nodule formation and nitrogen fixation (reviewed in Oldroyd & Downie, 2008; Popp & Ott, 2011). Competitiveness and nitrogen fixation are controlled by specific genes that are expressed during different time points in nodule development (Ampe et al., 2003; Barnett et al., 2004). Despite early work suggesting that plants could select more beneficial rhizobia from a mixture of effective and ineffective strains, that is, more effective strains were more competitive for nodules (Robinson, 1969), experiments with near-isogenic mutants demonstrated that, in four-fifths of cases, an ineffective mutant had the same competitiveness as its parent (Amarger, 1981). Third, some ineffective strains are very competitive under field conditions. In the agronomic literature, the infection of cultivated legumes by local populations of ineffective rhizobia has been termed ‘the competition problem’– competitive local strains can cause yield decreases of legume crops (Triplett & Sadowsky, 1992). The potential for defection could be circumvented by evolutionary mechanisms that maintain cooperation.

Evolutionary theory posits several mechanisms with the potential to maintain cooperative behavior between species in the absence of parent–offspring transmission, but we are only beginning to understand the molecular basis of these mechanisms in the legume-rhizobium mutualism. Potential stabilizing evolutionary mechanisms for the maintenance of symbiont traits that benefit the host are ‘partner choice’ or ‘screening’, in which hosts interact preferentially with partners that provide greater benefit, and ‘host sanctions’ or ‘partner fidelity feedback’, wherein hosts withhold resources from unproductive symbionts (Archetti et al., 2011). In partner choice, signals produced by the symbiont are used by the host to decide with whom to interact (Noë & Hammerstein, 1994); in screening, hosts impose costs for the symbiont to engage in the interaction and thus symbionts self-select (Archetti et al., 2011). In both cases, hosts interact more often with better strains. Both legumes and rhizobia produce signal molecules in the dialogue leading to infection, consistent with observations of partner choice (Heath & Tiffin, 2009; Gubry-Rangin et al., 2010). Furthermore, there are many examples of plants altering the composition of their rhizosphere communities (reviewed in Savka et al., 2002), suggesting the possibility of screening. After infection, partners exchange resources – plants provide rhizobia with sugars derived from photosynthesis and rhizobia use some of the energy from these sugars to drive the ATP-costly nitrogen fixation reaction that converts atmospheric nitrogen to ammonium, which is then exported to the host either directly or in the form of amino acids (Prell & Poole, 2006). Two reciprocity mechanisms could enforce the fair trade of resources: host sanctions and partner fidelity feedback. In partner fidelity feedback, hosts respond to the amount of resources provided by the symbiont by adjusting their allocation (Bull & Rice, 1991; Sachs et al., 2004). Host sanctions, in the narrow sense, are differentiated from partner fidelity feedback by hosts reducing resources below the partner fidelity feedback response level in order to select against cheating strains (Weyl et al., 2010; Archetti et al., 2011). A broader definition of host sanctions encompasses partner fidelity feedback and is defined as any reduction in host input when symbiont output drops (Kiers & Denison, 2008). There is evidence that legumes indeed regulate nodule allocation and symbiont reproduction in response to the rhizobium nitrogen fixation rate (Kiers et al., 2003, 2006), but we do not know whether this occurs beyond the level expected under the host response to nitrogen per se. The physiological coupling of host and symbiont fitness is shown in Fig. 1; there are two types of opportunity for antagonistic coevolution, based on infection rate and resource exchange, which are described in the following paragraphs.

Figure 1.

Simplified diagram of processes underlying legume and rhizobium fitness alignment and conflict. Processes (ovals) and relationships (arrows) are separated into those that represent direct feedback to the relative fitness of a strain (red) vs feedbacks that influence absolute fitness of all strains infecting a plant (blue; this could include bacterial cell number and cell quality as long as benefits are distributed equally across strains). Rectangles are observable quantities; bold rectangles represent the most commonly measured aspects of the interaction: aboveground biomass, nodule biomass, nodule number and relative nodule number. Signals modulate specific processes, such as phytohormones altering plant allocation patterns. Signals that influence infection thread formation (reviewed in Gage, 2004) include rhizobium-produced nod factor (reviewed in Long, 1996), succinoglycan (Jones et al., 2008) and plant-produced molecules that regulate root hair infection, including CLE peptides, ethylene, jasmonic acid and abscisic acid (Penmetsa & Cook, 1997; Ding et al., 2008; Ferguson et al., 2010; Batut et al., 2011). Many of these plant-derived molecules are known to act both locally and systemically (e.g. Mortier et al., 2010). Several plant genes required for infection thread development have been identified (Popp & Ott, 2011), and both nod factor (Walker & Downie, 2000) and bacterial surface molecules (lipopolysaccharide (LPS) and succinoglycan) can regulate infection thread abortion (Dazzo et al., 1991; Cheng & Walker, 1998). PHB, polyhydroxybutyrate.

Many more rhizobia inhabit a plant’s rhizosphere than the number of nodules formed, causing competition for nodulation and the potential for partner choice and/or screening mechanisms. Nodulation competition is poorly understood at the molecular level, compared with our knowledge of the molecular basis of nodule development in individual nodules (Jones et al., 2007; Oldroyd & Downie, 2008). Several rhizobium properties have been linked to nodulation competitiveness (Fig. 1): the alteration in plant hormone levels, particularly auxin and ethylene; the metabolism of complex molecules, such as amino acids and organic acids (Wielbo et al., 2007) as well as rhizopines (Murphy et al., 1995; Toro, 1996; Ratcliff & Denison, 2009); the production of antibiotic molecules, for example, trifolitoxin (Robleto et al., 1998); and motility and adhesion (Malek, 1992; Lodeiro et al., 2000). Legumes control the number of nodules formed through ‘autoregulation’ by a shoot-derived signal that limits infection (Ferguson et al., 2010; Reid et al., 2011), as well as via the phytohormone ethylene, as shown by ethylene-insensitive supernodulating mutants (Penmetsa & Cook, 1997; Penmetsa et al., 2003). However, rhizobia can influence these mechanisms by altering ethylene levels (Ma et al., 2002, 2004; Sugawara et al., 2006). Host regulation of nodule number may play a role in rhizobium competition for nodules through local or systemic regulation. Fig. 1 shows two examples of known mechanisms that influence the formation of infection threads in root hairs and infection thread abortion. Ethylene is inferred to act both systemically as well as locally, as demonstrated by the increased relative nodule number of strains that lower ethylene levels (Ma et al., 2004). Similarly, rhizobium strains with mutated cell-surface lipopolysaccharide (LPS) molecules have reduced nodulation competitiveness against their wild-type parent and are impaired in their ability to induce the systemic host autoregulation response in split root experiments (Lagares et al., 1992); furthermore, treatment with incompatible LPS causes infection thread abortion and plant defense responses (Dazzo et al., 1991). Thus, nodulation competitiveness and nodule number are influenced by multiple rhizobium characteristics.

Once inside a nodule, rhizobium reproduction and access to host resources are determined, in part, by how much nitrogen they fix, as legumes regulate allocation to nodules (Kiers et al., 2003; Sachs et al., 2010b), but should also depend on the overall health of their host, which should be influenced by all of the rhizobium strains that occur within nodules (Denison, 2000). The molecular basis of this regulation is currently unknown, but, in soy (Glycine max), it has been shown that a decrease in nitrogen fixation by the replacement of atmospheric nitrogen with argon triggers a reduction in oxygen supply to the nodule interior (Kiers et al., 2003). Rhizobia face a trade-off between the use of plant-provided sugars to fuel nitrogen fixation and the hoarding of these resources in storage molecules, such as polyhydroxybutyrate (PHB), which increases future survival and reproduction (Ratcliff et al., 2008; Fig. 1). The molecular nature of this trade-off has been confirmed with Rhizobium etli PHB synthase mutants and Rhizobium tropici glycogen synthase mutants, which both show increased rates of nitrogen fixation (Cevallos et al., 1996; Marroqui et al., 2001). However, in other bacteria, there are examples of mutations that cause positive covariation between storage and nitrogen fixation: in Azotobacter vinelandii, which is not a legume symbiont, mutation of phosphoenolpyruvate-protein phosphotransferase (ptsP) reduces both PHB hoarding and the nitrogen fixation rate (Segura & Espin, 1998). Plants with more beneficial rhizobium symbionts should have more total resources, which could feed back into more nodulation sites and more resources to allocate to nodules (Fig. 1). These represent common goods unless they are preferentially directed towards more beneficial strains, as shown for nodule biomass (Kiers et al., 2003; Oono et al., 2011). However, plants may not be able to discriminate between multiple strains that coinfect a single nodule. Despite these potential avenues for exploitation and the suggestion that some rhizobium strains that contribute little benefit to their hosts are parasites (Denison, 2000; Denison et al., 2003), we currently lack examples in which less beneficial rhizobium strains have higher fitness than more beneficial strains.

These mechanisms raise the question of whether the model legume–rhizobium symbiosis is undergoing antagonistic coevolution, where ‘cheating’ symbiont mutations that weaken or evade host control mechanisms readily evolve, or mutualistic coevolution, where hosts select for increasingly beneficial symbionts. Antagonistic coevolution predicts that the correlation between host fitness and symbiont fitness is negative. If mechanisms align partners’ fitness interests overall, ‘cheating’ rhizobia could still exist and be detected as outliers with high fitness that cause low host fitness. Two types of experiment can shed light on fitness correlations: single-strain inoculation and competitive inoculation. Plant performance with single-strain inoculation is a direct estimate of a symbiont’s potential impact on host fitness, but nodule size and number in these experiments conflate direct and indirect symbiont fitness benefits. Direct fitness benefits increase a strain’s relative fitness within the population, whereas indirect fitness benefits increase the absolute fitness of all strains interacting with a host. Direct symbiont benefits must be measured in mixed inoculation experiments – nodulation competitiveness is the most commonly measured trait, quantified through either inoculation with strain mixtures or by competing a strain against natural strains present in soil. Unfortunately, although host sanctions/partner fidelity feedback may play a major role in maintaining the legume–rhizobium mutualism (Kiers et al., 2003; Kiers & Denison, 2008), methodological challenges have thus far limited its empirical estimation across natural populations.

To test the hypothesis that legumes and rhizobia are undergoing antagonistic coevolution, the pattern of legume–rhizobium fitness correlation was assessed in two independent ways. First, studies were identified that measured the effects of individual rhizobium mutations on nodulation competitiveness and host performance. These studies enabled the estimation of between-species pleiotropy and the overall mutation pressure (i.e. the mutation rate multiplied by the mutational target size) of different types of rhizobium mutations. Second, studies were identified of natural rhizobia strains that measured rhizobium and legume fitness components, and meta-analysis was used to calculate pooled correlation coefficients between host and symbiont fitness. The results highlight the under-studied aspects of the legume–rhizobium symbiosis and fruitful directions for future research.

Materials and Methods

Literature survey

This study made use of the extensive collection of agronomic literature available through the University of California, Davis, CA, USA. ISI’s Web of Science was searched in June 2008 with ‘TI = (rhiz* OR bradyrhiz* OR mesorhiz* OR sinorhiz*) AND TI = (effective* OR efficien* OR nitrogen-fix* OR interaction*)’. This returned c. 1660 records, which were visually inspected for titles regarding plant performance. The resulting c. 940 papers were searched for ‘competit*’ to obtain c. 130 papers on competitiveness and effectiveness. Each was skimmed to determine whether the study investigated mutants, natural strains or both. In addition, any paper cited as having mentioned competitiveness and effectiveness was looked up. In August 2011, Google Scholar was used to find papers with ‘rhizobia’‘competition’‘effectiveness’ since 2007. Papers were divided into those that investigated spontaneous or induced rhizobium mutations (Table 1), or natural variation in rhizobium strains (Table 3). Mean values for individual natural strains were taken from tables or extracted from digital figures using DataThief (Tummers, 2006).

Table 1.   Studies used to determine pleiotropic effects of rhizobium mutations
StudyPlantRhizobiaN1Gene/pathwayMutation typeRatio2
  1. Rhizobium genus abbreviations: B., Bradyrhizobium; M., Mesorhizobium; R., Rhizobium; S., Sinorhizobium (now Ensifer (Willems et al., 2003)).

  2. 1Number of mutant strains considered.

  3. 2Inoculum ratio (mutant : reference) or number of ratios tested.

Amarger (1981)Medicago sativaS. meliloti6Antibiotic resistance: viomycin (1), kanamycin (1), neomycin (4)Spontaneous3
Brockman et al. (1991)Pisum sativum/Phaseolus vulgarisR. leguminosarum18UnknownTn5 insertion (10) or pGS9 integration (8)1 : 1
Bromfield & Jones (1979)Trifolium repensR. trifolii5Antibiotic resistance: streptomycin spectinomycinSpontaneous10 : 1
Date & Hurse (1992)Desmodium intortumBradyrhizobium6Antibiotic resistance: rifampicinSpontaneous1 : 1
Jensen et al. (2002, 2005))Medicago sativaS. meliloti3thuA, thuB (trehalose catabolism) and thuE (trehalose uptake)Insertional inactivation2–4
Jiang et al. (2001)Glycine maxS. fredii1idhA (myo-inositol dehydrogenase)Insertional inactivation5
Jiménez-Zurdo et al. (1995)Medicago sativaS. meliloti1proDH (proline dehydrogenase)Tn5 insertion1 : 1
Kuykendall et al. (1996)Glycine maxB. japonicum1Tryptophan pathway (gene unknown)Spontaneous2
Ma et al. (2004)Medicago sativaS. meliloti1acdS & lrpL (ACC deaminase)Insertion of gene into novel strains2
Malek (1992)Medicago sativaS. meliloti1Motility (gene unknown)Tn5 insertion4
Olah et al. (2001)Medicago sativaS. meliloti1ntrR (homologous to virulence proteins)Tn5 insertion1 : 1
Onishchuk et al. (2001)Medicago sativaS. meliloti6UnknownTn5 insertion1 : 1
Pankhurst (1981)Lotus pedunculatusM. loti4Antibiotic resistance: spectinomycin (1), d-cycloserine (3)Spontaneous1 : 1
Pankhurst et al. (1986)Lotus pedunculatusM. loti1Plasmid lossSpontaneous1 : 1
Parniske et al. (1993)Glycine maxB. japonicum1exoBTargeted deletion3
Rojas-Jiménez et al. (2005)Phaseolus vulgarisR. tropici2sycA (CIC-Cl-channel) and olsC (membrane lipid modifications, downstream of sycA)Targeted deletion10 : 1
Sessitsch et al. (1997)Phaseolus vulgarisR. tropici5UnknownTn5 gusA insertion5
Triplett (1990)TrifoliumR. leguminosarum bv. trifolii1tfx genes (trifolitoxin)Insertion of gene into novel strain3
Turco et al. (1986)Pisum sativum/Glycine maxR. leguminosarum/B. japonicum16Antibiotic resistance: streptomycin, kanamycin, rifamycin, erythromycin, spectinomycinSpontaneous3

Mutational variation

There were 19 papers found that measured both nodulation competitiveness and effectiveness of mutant rhizobia, with competition experiments performed either between strains that differed from one another by single mutations (i.e. a mutant strain competed against its parent) or strains differing from one another by a single mutation competed against a common reference strain. In one study (Amarger, 1981), strains differed from one another by two mutations, but additional competition experiments were performed to ascribe the effect to the mutation causing ineffective symbiosis rather than the marker mutation. The species used and types of mutations investigated are presented in Table 1. These studies gave a total of 80 mutant comparisons, summarized in Table 2. Each mutation was classified as increasing, decreasing or not significantly affecting rhizobium nodulation competitiveness and host benefit as measured and reported in the original paper. The binary nature of much of the primary data precluded formal meta-analysis of fitness correlations arising from pleiotropy. It is known that rhizobium competitive ability is often frequency dependent (Amarger & Lobreau, 1982), but, as the competition coefficient is defined as the intercept of the regression of log(nodulation ratio) on log(inoculum ratio), this is equivalent to assessing competitive ability at a 1 : 1 inoculum ratio (log(1) = 0). Furthermore, these types of study do not differ in their pattern of rhizobium fitness (Fisher’s exact test, = 0.9). The two studies that only performed competition at 10 : 1 ratios of mutant : reference are problematic, as these data are not equivalent to the other 17 studies. However, these studies do not have a different pattern of rhizobium fitness from the others (Fisher’s exact test, = 0.33).

Table 2.   Summary of pleiotropic effects of rhizobium mutations on legume and rhizobium fitness
 Rhizobium fitness
Plant fitness

Natural variation

Surveying papers that measured components of host and symbiont fitness in naturally occurring rhizobium strains identified 26 studies with 34 experiments (see Table 3). Only seven experiments were conducted on plants that made indeterminate nodules. Although the search was extensive for studies that measured both nodulation competitive ability, that is, relative fitness, as well as host benefit, the search for papers measuring the rhizobium absolute fitness components, that is, nodule number and nodule biomass, was not nearly exhaustive. It would be interesting in the future to assemble these data to determine whether there are different patterns of growth correlations among strains that associate with legumes which form different types of nodule (e.g. indeterminate vs determinate) or among strains sampled from different geographic locations (e.g. latitude), environmental conditions (e.g. soil nitrogen content) or spatial scales. Such an effort is beyond the scope of this review. Of the studies compiled, several tested the same strains on multiple host species or genotypes, and several tested different categories of strains (e.g. isolated from different hosts, fast vs slow growing) on the same host. When the same set of strains was tested for competitiveness and effectiveness on multiple cultivars or hosts, these were considered as separate experiments. An argument in favour of this approach is that, in some cases, the relationship between host and symbiont was dependent on the host being tested. Only studies reporting data on at least five strains were included. Table 3 summarizes the phenotypes measured, the species studied and the growth conditions of each experiment.

Table 3.   Studies used in meta-analysis of natural strains
Study1Nod2Plant speciesRhizobiaNPlant phenotypes3Rhizobium phenotypes4Cond5
  1. 1Studies: AME89, Ames-Gottfred & Christie (1989); BROWN96, Brown & Ahmad (1996); CHA85, Chandra & Pareek (1985); CHEN02, Chen et al. (2002); DATE92, Date & Hurse (1992); FUHR89, Fuhrmann & Wollum (1989); HAF01, Hafeez et al. (2001); HUN00, Hungria et al. (2000); HUN01, Hungria et al. (2001); HUN03, Hungria et al. (2003); HUN98, Hungria et al. (1998); KUN08, Kundu & Dudeja (2008); LAG07, Laguerre et al. (2007); NUR88-89, Nurhayati et al. (1988, 1989); PAN79, Pankhurst & Jones (1979); PAN81, Pankhurst (1981); RICE95, Rice et al. (1995); ROD99, Rodriguez-Navarro et al. (1999); SACHS10, Sachs et al. (2010a,b); SAX96, Saxena et al. (1996). SEN81, Sen & Weaver (1981); TAM02, Tamimi (2002); THI04, Thiao et al. (2004); TRI89, Trinick & Hadobas (1989); TRI90, Trinick & Hadobas (1990); VAR07, Vargas et al. (2007).

  2. 2Nod: nodule types: D, determinate; I, indeterminate.

  3. 3Plant phenotypes: ABM, aboveground biomass; ARA, acetylene reduction assay (measure of nitrogen fixation activity); GRAIN, total weight of seeds; HI, harvest index; Ht, plant height; Nit, nitrogen content; Nit_TOT, total nitrogen content of plant; Pod_n, total number of pods; Pod_wt, weight of pods; RBM, root biomass; score, performance score; Stem_d, stem diameter.

  4. 4Rhizobium phenotypes: B_prot, bacteroid protein; cfu, total colony-forming units recovered from nodules of a plant; Comp, competition relative to several other strains (averaged); Comp vs ref, competition coefficient relative to a standard reference strains; Comp vs soil, competition measure against unknown soil population; N, nodule number; NBM, total nodule biomass per plant; Nod_nit, total nitrogen content of nodules.

  5. 5Cond: growth conditions: A, sterile conditions; F, field; G, glasshouse; –, not reported.

AME89ITrifolium pratenseR. trifolii10ABM; NitComp vs soilA, F
BROWN96DPhaseolus vulgarisR. tropici6ABMN; CompA
CHA85ICicer arietinum (G-114)Rhizobium5ABM; GRAINComp vs soilF
CHA85ICicer arietinum (JG-62)Rhizobium5ABM; GRAINComp vs soilF
CHEN02DGlycine maxVarious81ABM; NitN; NBMA
DATE92DDesmodium intortumBradyrhizobium6ABMComp vs refA
FUHR89DGlycine maxBradyrhizobium5ABM; NitN; NBM; CompA
HAF01DVigna radiataBradyrhizobium15ABM; NitN; NBM; Comp vs refG
HUN00DPhaseolus vulgarisRhizobium8NitN; NBM; Comp vs refA
HUN01DGlycine max (BR-16)Various32ABM; NitN; NBM; Comp vs refA
HUN01DGlycine max (Davis)Various32ABM; NitN; NBM; Comp vs refA
HUN03DPhaseolus vulgarisR. tropici7NitN; NBMF
HUN98DGlycine maxB. japonicum, B. elkanii38NitN; NBM; Comp vs refA
KUN08DVigna radiataRhizobium10ABM; RBM; Nit_TOTN; NBM; Comp vs soilG
LAG07IPisum sativumR. leguminosarum bv. viceae42ABM; RBM; NitN; NBMG
NUR88-89DCentrosema pubescensVarious6ABM; RBM; NitNBM; Comp vs soilA, F
NUR88-89DDesmodium heterocarponVarious6ABM; RBM; NitNBM; Comp vs soilA, F
NUR88-89DMacroptilium atropurpureumVarious6ABM; RBM; NitNBM; Comp vs soilA, F
NUR88-89DVigna unguiculataVarious6ABM; RBM; NitNBM; Comp vs soilA, F
PAN79DLotus pedunculatusRhizobium6ABM; Nit; ARANBMA
PAN81DLotus pedunculatusRhizobium7ScoreComp vs refA
RICE95IMedicago sativaS. meliloti20ABMN; NBMA
ROD99DPhaseolus vulgarisR. etli, R. leguminosarum15ABM; Pod_wt; Pod_n; HIN; NBMA
SACHS10DLotus strigosusBradyrhizobium5ABMN; NBM; cfu; CompA
SAX96DVigna radiataBradyrhizobium10ABM; RBM; ARA; GRAINN; NBM; Comp vs soilG, F
SEN81DArachis hypogaeaRhizobium6ABM; ARA; NitNBM; B_prot
SEN81DMacroptilium atropurpureumRhizobium6ABM; ARA; NitNBM; B_prot
SEN81DVigna unguiculataRhizobium6ABM; ARA; NitNBM; B_prot
TAM02DPhaseolus vulgarisRhizobium10ABM; NitN; NBMA
THI04IGliricidia sepiumRhizobium, Bradyrhizobium10ABM; RBM; NitNBMA
TRI89DMacroptilium atropurpureumBradyrhizobium17ABM; NitN; NBM; Comp vs refA
TRI89DVigna unguiculataBradyrhizobium26ABM; NitN; NBMA
TRI90I/DParasponia andersoniiBradyrhizobium20ABM; NitNBM; Nod_nitA
VAR07IAcacia mearnsiiBradyrhizobium10ABM; Nit; Ht; Stem_dNBMA

Two measures of rhizobium relative fitness were used: competition coefficients and nodule occupancy, that is, relative nodule number (Fig. 1). Competition coefficients are defined as the logarithm of the ratio of the strains in the nodules to the logarithm of the ratio of the strains in the inoculum; these were estimated for studies that competed strains against a reference strain. For studies that introduced marked strains into a natural soil, the resulting percentage nodule occupancy was used as an estimate of competitive ability. Finally, some studies performed multiple pairwise competition assays and reported either competition coefficients or nodule occupancy. Multiple trials were averaged across and the original fitness measure was retained. Only one study, Fuhrmann & Wollum (1989), conducted competition assays across a range of inoculum ratios. For this study, the competition coefficients (i.e. the intercept of the regression of log(nodulation ratio) on log(inoculum ratio)) were used. Analysis was conducted on two measures of rhizobium absolute fitness that are commonly reported: nodule biomass and nodule number in single-strain inoculations (Fig. 1). These were considered separately from the relative fitness measures. Aboveground biomass is the most commonly reported plant fitness component (Fig. 1, bold rectangle).

The Pearson correlation coefficient was calculated between traits of interest, as well as the probability that each correlation coefficient is 0, using a custom script in R version 2.13.1 (R Development Core Team, 2011); code and raw data are available on request. One study did not report data for each strain, but, rather, presented overall correlation coefficients, which were used directly (Laguerre et al., 2007). Independent experiments from each study were pooled using Fisher’s Z-transformation, = (1/2) loge((1 + ri)/(1 − ri)), with inverse variance weighting, wi = Ni − 3, where ri is the Pearson correlation coefficient and Ni is the sample size in the ith study (DerSimonian & Laird, 1986; Fagard et al., 1995; Moher et al., 2000). Outliers were identified using the R package ‘mvoutlier’ (Filzmoser et al., 2012) with the minimum covariance determinant (MCD) parameter estimated with 0.8 of the data and a chi-squared cutoff of 1%. Outliers were tabulated with respect to their position relative to the mean plant and bacterial fitness components; full results are plotted in Supporting Information Figs S1–S4. Differences across each table were tested with Fisher’s exact test and deviations from the null hypothesis in each row and column were tested using the binomial distribution function. For this analysis, as there are five tests performed per table and four tables (20 tests in total), Bonferroni correction adjusts the significance threshold to 0.05/20 = 0.0025.


Rhizobium mutations tend to pleiotropically align legume and rhizobium fitness

Surveying 80 rhizobium mutants from 19 papers shows that pleiotropy most often aligns host and symbiont fitness (Table 2). Notably, the only two cases of antagonistic pleiotropy are mutations that are detrimental to rhizobium competitive ability, yet increase host performance. There are no cases of ‘cheater’ mutations in which increased rhizobium fitness occurs with a pleiotropic decrease in host fitness. Indeed, there is a general pattern of positive fitness correlations: rhizobia with mutations deleterious to their own competitive ability tend (19/44) to have lower effectiveness (−, −), whereas four of 12 mutations that increase competitive ability increase effectiveness (+, +). Considering only mutations that have a significant effect on both partners, there is significant nonuniformity in the contingency table (Fisher’s exact test: = 0.0012, = 25). Furthermore, mutations that decrease plant fitness are significantly more likely to be detrimental for rhizobia (= 1.9e-06, = 19), and mutations that decrease rhizobium fitness are significantly more likely to be detrimental to their plant host (= 0.00011, = 21). Positive pleiotropy suggests that mutations enabling rhizobia to increase their fitness at the expense of their host are rare relative to those mutations that are either deleterious or beneficial to both partners.

Natural variation in rhizobia and legume fitness components

Nodulation competitiveness, measured as the relative nodule number (Fig. 1), is a rhizobium fitness component that indicates directly the relative fitness of different strains. Across 19 studies, the pooled correlation between rhizobium nodulation competitiveness and plant performance is significantly positive (= 0.31, 95% CI = (0.13, 0.46), = 19). For studies that competed each strain against a reference strain, the pooled correlation is 0.65 (Fig. 2a, 95% CI = (0.31, 0.84), = 5) and, for studies that competed each strain against an uncharacterized population of rhizobia in nonsterile soil, the pooled correlation is 0.23 (Fig. 2b, 95% CI = (0.041, 0.401), = 14). These pooled correlation coefficients include two studies with negative fitness correlations. One of these studies’ negative correlations is driven by a single strain (NUR88-89), but the other is robust to the exclusion of a beneficial strain that is an extremely poor competitor – the remaining five strains still produce a correlation of −0.90 with = 0.0381 (DATE92). It is intriguing that this is one of only two studies that measured competitiveness on the same host from which the strains were isolated.

Figure 2.

Forest diagram showing meta-analysis of correlations between relative (competitive) fitness measures of rhizobia with legume performance. Each study mean is indicated by a box whose size is scaled proportional to the number of rhizobium strains used. The 95% confidence interval of each study is indicated by the horizontal line extending from the box. The pooled correlation coefficient and its 95% confidence interval are shown by the blue diamond. The null hypothesis of no correlation is indicated by a gray vertical broken line. (a) Rhizobium fitness metric is the competition coefficient estimated by competing strains against a reference; plant fitness metric is the aboveground biomass with each strain in single-strain inoculation. (b) Rhizobium fitness metric is the nodule occupancy (percent) measured by competing strains against an unknown soil population or against multiple other strains; plant fitness metric is the aboveground biomass with each strain in single-strain inoculation.

Nodule number in single-strain inoculation studies is not a measure of rhizobium relative fitness, as ineffective strains typically produce a large number of nodules when inoculated alone, but are not necessarily more competitive than effective strains when inoculated together (e.g. Oono et al., 2009; Gubry-Rangin et al., 2010; Fig. 1). With this caveat in mind, meta-analysis of 20 experiments found the pooled estimate of the correlation between nodule number and plant aerial biomass to be 0.76 (Fig. 3a, 95% CI = (0.73, 0.84), = 20). No individual studies showed a significant negative overall correlation. Similarly, although nodule biomass in single-strain inoculations is an estimator of potential rhizobium absolute fitness, it is not an appropriate measure of rhizobium relative fitness, as larger plants probably have greater resources overall which could reflect a ‘common good’ that may be exploited. Across 28 experiments that measured nodule biomass and aerial biomass, the overall pooled correlation is 0.83 (Fig. 3b, 95% CI = (0.79, 0.86), = 28). One study found a significant negative correlation for ‘big-nodule’ strains on the Pisum sativum cultivar Austin, but not on cultivar Frisson or across ‘small-nodule’ strains. Nodule size is sometimes (Heath & Tiffin, 2007), but not always (Wood et al., 1985; Sachs et al., 2010a), correlated with the number of viable rhizobia inside a nodule.

Figure 3.

Forest diagram showing meta-analysis of correlations between absolute (noncompetitive) fitness measures of rhizobia and legume performance. Figure components as in Fig. 2. (a) Rhizobium fitness metric is the nodule biomass in single-strain inoculations; plant fitness metric is the aboveground biomass. (b) Rhizobium fitness metric is the nodule number in single-strain inoculations; plant fitness metric is the aboveground biomass.

Overall, studies of naturally varying strains show predominantly positive fitness correlations between rhizobia and legumes for both relative and absolute fitness components of rhizobia.

Outlier analysis

Different rhizobium strains may use different molecular pathways to interact with their legume hosts. If a minority of the strains sampled possess novel mechanisms of interaction that influence the relationship between rhizobium fitness and plant fitness, they will appear as outliers in these distributions. As they are outliers, the null hypothesis is that these strains are equally likely to fall above and below the mean values for both partners’ fitness. Cheating rhizobium strains would have higher than average fitness, but yield lower than average plant fitness, whereas, if rhizobia are being exploited by their hosts, they would have lower than average fitness, but yield higher than average plant fitness. The number of outliers in each quadrant is given in Table 4.

Table 4.   Summary of outliers in studies of natural rhizobium strain variation
 R−R+Row P
  1. R−, rhizobium fitness component less than mean; R+, rhizobium fitness component greater than mean. P−, plant performance less than mean; P+, plant performance greater than mean. Col. (column) P and Row P, probability of observing the given configuration, or one more extreme, under the null hypothesis that outliers are equally likely to deviate in the positive and negative direction.

  2. Bold type, P value of Fisher’s exact test on the contingency table.

Rhizobium competition coefficient
 Col. P0.750.51.00000
Rhizobium nodule occupancy
 Col. P0.50.271.00000
Rhizobium nodule number
 Col. P0.06250.180.07304
Rhizobium nodule biomass
 Col. P0.051.11E-040.00011

An alternative definition of a ‘cheating’ outlier would be a strain that provides less than the expected amount of host benefit, given the rhizobium fitness obtained. In Figs S1–S4, these would be outliers that fall below the overall data. For both relative fitness measures and one absolute fitness measure, there is no deviation from the null expectation of half the outliers falling below the overall data cloud and half falling above (competition coefficient: 1 vs 2, nodule occupancy 6 vs 6, nodule number 14 vs 12). For nodule biomass, there are significantly more strains that are outliers above the data cloud than below (27 above vs 7 below, = 0.00041). However, as the fitness of each partner is not zero-sum, it is not meaningful to call strains that provide less than the expected host fitness for their fitness ‘cheaters’.

There are few outliers for the relative fitness measures nodulation competitiveness and nodule occupancy, and neither show deviations from the null hypothesis (Table 4). Rhizobium strains that yield higher than average plant fitness have a tendency to also have higher than average nodule occupancy, but it is not significant after Bonferroni correction (= 0.035, Bonferroni threshold for 20 tests = 0.0025). For absolute rhizobium fitness measures, nodule number does not show an overall difference from the null hypothesis (Fisher’s exact test: = 0.073), but nodule biomass does (Fisher’s exact test: = 0.000113). Rhizobia that yield higher plant fitness also have significantly higher nodule number (= 0.00171) and nodule biomass (= 0.000427). Similarly, strains that have higher than average nodule biomass yield higher than average plant growth (= 0.000110). Thus, outliers point to variation in mechanisms that even more tightly couple legume and rhizobium fitness interests.


Legume and rhizobium fitness interests are aligned

Despite the hypothesis that cheaters pose a destabilizing threat to the legume-rhizobium mutualism, we have little evidence for their existence. Under antagonistic coevolution, we expect to observe fitness conflict between partners. Mechanisms that align host and symbiont fitness, depicted in Fig. 1, can prevent antagonism and instead lead to mutualistic coevolution, where partners evolve to be increasingly beneficial to one another (Law, 1985). Ancient fitness conflict could have selected for such fitness aligning mechanisms (West et al., 2002), or these mechanisms could have arisen for other reasons, such as growth allocation in response to general nutrient conditions, and thus represent ‘exaptations’ (Gould & Vrba, 1982) that enabled the origin of the mutualism. If fitness conflict was the predominant force shaping the legume-rhizobium interaction, we would expect to observe overall negative fitness correlations between partners. In fact, we observe the opposite: both individual mutations and natural variation in rhizobium strains reveal a pattern of fitness alignment.

Studies of rhizobium mutations strongly support the hypothesis of fitness alignment. The legume–rhizobium interaction appears to be structured so that mutations beneficial for the symbiont have pleiotropic benefits on their host. Conversely, most mutations that are deleterious to the host are similarly deleterious to symbiont competitive ability. This positive pleiotropy suggests that many ineffective strains in nature are ‘defective’ rather than ‘defectors’. Mutations that increase rhizobium competitive ability either increase or do not impact on host performance. In fact, the only examples of fitness conflict are mutations that are costly to rhizobium fitness, but benefit the plant. These could only evolve under extreme evolutionary scenarios, such as enslavement (Frean & Abraham, 2004). There are no examples in the literature surveyed here of ‘cheater’ mutations that increase rhizobium fitness at the host’s expense. As most mutations are weakly deleterious (reviewed by Lynch et al., 1999), it is not surprising that ‘defective’ rhizobia are less competitive and provide less host benefit, but this does not predict the mutually beneficial effects of some rhizobium mutations and the mutations that enhance host growth at the expense of rhizobium fitness. These studies are probably biased towards finding dramatic plant growth decreases when symbiosis is disrupted, rather than subtle plant growth increases. However, there is no corresponding directional bias in detecting altered nodulation competitive ability. It is interesting to note that two studies found decoupling of rhizosphere fitness and nodulation competitive ability: the trehalose catabolism thuB mutant of Sinorhizobium meliloti has poor rhizosphere colonization ability, but still infects more nodules than its parent (Jensen et al., 2005), and spontaneous mutants of S. meliloti show no correlation between competitiveness in the soil, rhizosphere or for nodulation (Onishchuk et al., 2001). Further work showed that rhizosphere and nodulation competitiveness depends on host genotype, and that competitiveness for nodules is correlated with the induction of thuB in infection threads (Ampomah et al., 2008). Although the survey presented here is a small sample of all possible mutations in rhizobia, especially given the variation in gene content between strains (Young et al., 2006; Galardini et al., 2011), it nonetheless shows that ‘cheater’ mutations are rare compared with mutations that have pleiotropic effects aligning host and symbiont fitness.

Mutant studies can uncover alleles that are rare in nature, such as the relatively common mutant alleles (c. 25%) that have deleterious effects on both host and symbiont fitness. Natural variation, however, also provides information on trait correlations that arise through linkage disequilibrium and epistasis, which are difficult to study using random mutations. Similar to the fitness alignment observed using single mutations, natural strains show significantly positive pooled correlations between host and symbiont fitness. Symbiont relative fitness, as measured by nodulation competitiveness, is positively correlated with host benefit in single-strain inoculations, and absolute measures of symbiont fitness in single-strain inoculations are even more positively correlated with host benefit. Some strains that are outliers have higher than average rhizobium fitness but cause lower than average plant fitness (Table 4). However, there are as many or more outlier strains that are ‘super-mutualists’ and fall above the mean for both host and symbiont fitness. Positive correlations between plant performance and nodule number/nodule biomass provide support for partner fidelity feedback, in which traits that enhance partner performance increase the partner’s ability to return benefits. When unrelated symbionts interact with a single host, these increased benefits could be exploited if they are a public good. However, if hosts reciprocate to particular symbionts based on their performance, often termed sanctions (Kiers et al., 2003), these benefits would be private goods and hence not exploitable.

Interpreting inoculation experiments as estimates of host and symbiont fitness

Most fitness assessments utilize single-strain inoculations, which conflate absolute and relative fitness (Oono et al., 2009). This problem is highlighted the case of a Californian strain of Bradyrhizobium that has high absolute fitness when inoculated alone – it forms many nodules with high bacterial cell numbers – but extremely low relative fitness in competition (Sachs et al., 2010a,b). Nodulation competitiveness is the only often-measured rhizobium fitness component that directly influences a strain’s relative fitness in a population (Fig. 1). The absolute fitness measures of nodule number and nodule biomass may represent public goods that are shared among the strains that infect a given host plant (Fig. 1); not enough is known about the regulating mechanisms to assess the degree to which these are public vs private goods. Plants may produce more and/or larger nodules because nitrogen fixed by effective strains allows increased plant growth, and less beneficial strains could potentially exploit these benefits. However, in some cases, the correlation between nodule number and plant performance in single-strain inoculations can be driven by a rhizobium mutation that causes increased relative fitness (Table 2). As more root hairs are infected than ultimately form nodules (Bauer, 1981), any rhizobium pathway that decreases infection thread abortion locally will have increased competitive ability, as well as making more nodules in single-strain inoculations (Fig. 1). Similar arguments can be proposed for the relationship between nodule size and plant performance, with additional support coming from studies of host sanctions that operate on individual nodules in response to nitrogen fixation (Kiers et al., 2003; Oono et al., 2011). We do not currently know whether legumes can apply sanctions within a nodule infected by a beneficial and ineffective strain; evolutionary modeling of this scenario suggests that, although it may lead to diversification into more and less beneficial strains, it is unlikely to explain the existence of fully ineffective rhizobia (Friesen & Mathias, 2010). Taken together, the positive correlations between rhizobium absolute fitness measures and plant performance may partly carry over to rhizobium relative fitness; thus, not all absolute fitness is ripe for exploitation.

Estimating host benefits from symbiosis is almost solely performed by inoculating plants with single strains. However, in nature, plants typically associate with many strains (McInnes et al., 2004) and both synergistic and antagonistic nonadditive effects have been observed with multistrain inoculation (Fuhrmann & Wollum, 1989; Heath & Tiffin, 2007). An ecologically motivated method for estimating the benefit conferred by a given rhizobium strain would be to vary its relative abundance against a diverse rhizobium community background and to regress plant performance on relative nodulation success. Although biomass has been shown to be broadly correlated with plant competitive ability for annuals (Gaudet & Keddy, 1988), it is not a direct measure of fitness. Biomass and/or seed production in the presence of a competitor, such as a non-nodulating mutant of the same plant species or non-nodulating natural competitor, would be a better host performance measure. Furthermore, reproductive output is not necessarily correlated with vegetative growth. In other symbioses, symbionts can delay host reproduction and increase vegetative growth (Baudoin, 1975; Clay, 1991). The potential conflict over reproductive timing has not been assessed in the rhizobia–legume symbiosis.

All but two of the studies synthesized here were performed with cultivars that do not grow naturally at the sites at which rhizobium strains were collected. These associations provide a null distribution of fitness effects in the absence of coevolution and indicate widespread potential for positive fitness correlations. Recent work in two natural systems provides additional evidence that these correlations are largely positive (included in this analysis: Sachs et al., 2010a,b; not included: Heath & Tiffin, 2009). Heath & Tiffin (2009) performed a multigenerational study in which natural genotypes of Medicago truncatula from France were replanted into soil in which they had grown previously in the presence of three wild French Sinorhizobium medicae strains. Nodule typing showed that the strain which gave the most plant benefit occupied an increasing fraction of nodules over each plant growth cycle.

The molecular basis of fitness aligning mechanisms

Mutations that enhance both partners’ fitness provide hypotheses for the molecular underpinnings of the fitness alignment described here. These genes are all members of pathways involved in signaling and, in two cases, may participate in the autoregulation of nodulation. First, a mutant of S. meliloti 1021 that nodulates Medicago sativa with ammonium present is altered in the ntrR gene, which has a conserved PIN domain involved in host–symbiont signaling (Olah et al., 2001). This strain increases host shoot biomass by 10% in the absence of nitrogen and occupies 100% of nodules when competed against its immediate ancestor. Nitrogen inhibition of nodulation is currently believed to share components of the autoregulation pathway (Reid et al., 2011). Second, a strain of S. meliloti engineered to contain a 1-aminocyclopropane-1-carboxylate (ACC) deaminase gene had a 20-fold competitive advantage and increased M. sativa biomass by 45% and nodule number by 35% compared with its ancestor (Ma et al., 2004). ACC deaminase lowers ethylene levels by reducing concentrations of ACC, the precursor to ethylene; this gene is found in many rhizobia and plant growth-promoting rhizobacteria (Glick et al., 2007). Ma et al. (2004) hypothesized that rhizobium ACC deaminase lowers ethylene levels and prevents infection thread abortion in the early stages of infection (e.g. Fig. 1). Finally, a revertant of a tryptophan mutant Bradyrhizobium japonicum strain increases plant nitrogen content by 50%, increases shoot dry weight by 31%, increases nodule number by 56% and increases nodule biomass by 42% (Hunter & Kuykendall, 1990). A follow-up study competed this strain against two different reference strains and found a 2.5- or six-fold higher competitive ability compared to its parent; this study also found increased soy yield under field conditions (Kuykendall et al., 1996). The precise mutation in this strain is not known, but alters tryptophan metabolism so that it is more sensitive to tryptophan. The tryptophan pathway is involved in the production of indole-3-acetic acid (IAA), that is, the phytohormone auxin. In S. meliloti, the addition of genes that increase auxin production causes higher host biomass (Bianco & Defez, 2009). In Rhizobium leguminosarum bv. viceae, overproduction of IAA increased seed number by 20% and caused increased nitrogen fixation rates and higher nodule biomass in single-strain inoculations (Camerini et al., 2008).

The pattern of gene-for-gene structured host restriction of nodulation observed in the legume–rhizobium interaction is indicative of antagonistic coevolution and suggests conflict over infection rate and nodule development. Well-known examples implicate variation in the structure of the nod factor signal that interacts with host recognition pathways (Triplett & Sadowsky, 1992; Firmin et al., 1993). A recent study in M. truncatula showed a negative genetic correlation between nodule number and fruit number (Heath, 2010). This study found no significant correlation between nodule number and leaf number, further emphasizing the need for studies that more accurately estimate host fitness. In the wild legume Amphicarpea bracteata, plant genotypes with restrictive nodulation, that is, extreme preferences, obtain greater benefits with their preferred ‘specialist’ rhizobia (Wilkinson et al., 1996). We require substantially more experiments to address whether there is a general relationship between host range and symbiotic effectiveness. However, although nodule number could be subject to conflict, the evidence compiled here implies the opposite. In two of the four cases in which a rhizobium mutation increased both partners’ fitness, there was an increase in the number of nodules formed in single-strain inoculations. These cases demonstrate that the host regulation of nodule number is sometimes suboptimal. One hypothesis for this pattern is that plants could be adapted to interacting with many microbes simultaneously. Plants might then overly restrict rhizobium symbionts when grown in sterile conditions, similar to the way in which animal immune systems require exposure to particular components of the microbiome for appropriate development (Mazmanian et al., 2005). Under this scenario, the correlation between rhizobium competitive fitness and plant performance should depend on the presence of natural microbial flora. Studies of natural rhizobia that perform competitive inoculations in otherwise sterile conditions show a higher pooled correlation between competitive ability and host performance than studies in which strains were competed against an uncharacterized soil community. This could arise from added variation as a result of soil heterogeneity or field conditions, or other soil microbes could alter the relationship between rhizobium competitiveness and effectiveness. Mycorrhizal symbiotic benefit is greater when co-inoculated with soil bacteria (Hoeksema et al., 2010); we require similar controlled studies with rhizobia and natural microbiota to measure ecologically meaningful interaction coefficients.

We know very little about the molecular basis of post-infection host regulation of rhizobia, that is, sanctions/partner fidelity feedback, beyond the fact that nodule oxygen permeability is involved (Kiers et al., 2003). The data considered here do not provide an assessment of sanctions in aligning legume and rhizobium fitness across naturally varying strains. This would require the apportioning of nodule biomass and viable rhizobium population size to each strain in competitive inoculations. Sanctions exist in determinate and indeterminate nodulating legumes (Oono et al., 2011), and restrict both nodule biomass allocation and rhizobial proliferation in response to nitrogen fixation (Kiers et al., 2003; Oono et al., 2009). At least two components are involved in this plant response, as there are two examples in which biomass is restricted, but rhizobia are able to maintain high population sizes within low-biomass nodules. One wild Bradyrhizobium strain provides no growth benefit to its Lotus strigosus host, but forms numerous small nodules with very high cell densities (Sachs et al., 2010a). Similarly, a Bradyrhizobium strain grown with a genotype of Glycine soja makes small nodules with degraded plant cells and large populations of live undifferentiated rhizobia, although the same strain is effective with G. max (Parniske et al., 1990).

Why are there ineffective rhizobia?

If mutualism-enhancing mechanisms, such as partner choice, screening, partner fidelity feedback and host sanctions, align the fitness of legumes and rhizobia, why are there rhizobia in nature that provide little host benefit? First, these rhizobia could be ‘defective’ as a result of deleterious mutations that decrease both host and symbiont fitness (Table 2). This could be tested in natural populations by measuring the effectiveness of many strains and the strength of host selection to determine whether the phenotype is segregating at mutation-selection balance. Second, ineffective strains could be conditionally defective, that is, perform poorly with some host genotypes or species (Turco et al., 1986; Parniske et al., 1990; Somasegaran et al., 1991; Wilkinson et al., 1996; Heath & Tiffin, 2007). Third, ineffective strains could represent soil or rhizosphere specialized genotypes that have recently acquired a gene island for symbiosis (e.g. Sullivan & Ronson, 1998; Sachs et al., 2010a). Finally, they may, in fact, be cheating the plant in a manner that has yet to be measured systematically.


Fitness alignment between legumes and rhizobia is key to the persistence of this mutualism, whether mediated by sanctions, screening, partner choice or partner fidelity feedback. A full understanding of this fitness alignment requires further studies to determine whether the positive fitness correlations synthesized here extend generally to natural coevolved associations. We also require assessment of the prevalence and ecological importance of fitness aligning mechanisms; in particular, we lack measurement of the degree of fitness conflict over nodule biomass in competitive inoculations and fitness conflict over other host traits, such as flowering time or root : shoot ratio. Furthermore, cultivar selection under high nitrogen fertilization regimes threatens to lose the plant genetic mechanisms that structure this beneficial interaction. More recent cultivars of soy are less able to maintain performance when interacting with a combination of high- and low-benefit rhizobia than are older cultivars (Kiers et al., 2007). We lack an assessment of the pleiotropic effects of plant mutations on their ability to align the interests of hosts and symbionts, and therefore do not know whether the change observed by Kiers et al. (2007) reflects the fact that these mechanisms impose a fitness cost to the host, or whether the decreased ability is an incidental effect of selecting on other traits. Regardless, the management of plant–microbe mutualisms for agriculture requires the elucidation of the mechanisms that align legume and rhizobium fitness interests. With recent advances in legume genomics, including the genome sequences of G. max and M. truncatula, together with the development of mapping populations and functional genomic tools, these mechanisms can now be sought. This will lead to a better understanding of host–symbiont coevolution and potentially enable the design of better symbioses for agriculture.


I am grateful to R. F. Denison and M. L. Stanton for incisive comments on an earlier draft. I thank R. F. Denison, W. C. Ratcliff, S. V. Nuzhdin and S. S. Porter for stimulating discussions and J. L. Sachs and K. D. Heath for sharing raw data. The comments of three anonymous reviewers improved the quality of this work. I acknowledge funding through the Natural Sciences and Engineering Research Council of Canada Post-Graduate Scholarship, Doctoral (NSERC PGSD) and the Center for Population Biology at the University of California, Davis, CA, USA.