Controlling the reproductive fate of rhizobia: how universal are legume sanctions?

Abstract

Contents

• Summary 967

• I. 
  Introduction 967
• II. 
  Evolutionary history: is host-imposed rhizobial dimorphism ancestral or derived in legumes? 970
• III. 
  Are there immediate benefits to individual plants from rhizobial dimorphism? 970
• IV. 
  How does dimorphism affect rhizobial evolution? 972
• V. 
  How might sanctions differ when bacteroids are nonreproductive? 974
• VI. 
  Conclusions 977

• Acknowledgements 978

• References 978

‘I use this term [struggle for existence] in a large and metaphorical sense including dependence of one being on another, and including (which is more important) not only the life of the individual, but success in leaving progeny.’

Darwin (1859)

I. Introduction

The legumes (Fabaceae) are a diverse family and their symbioses with nitrogen-fixing bacteria (rhizobia) may be equally diverse (Sprent, 2007). In particular, rhizobia are dimorphic (present in two different forms) in nodules of some legume hosts. In those symbioses, the bacteroids (the differentiated rhizobial cells that can fix nitrogen) are larger than their undifferentiated clonemates in the same nodule and have typically lost the ability to reproduce. By contrast, bacteroids in other hosts retain the ability to reproduce after fixing nitrogen, perhaps because they are more similar to undifferentiated rhizobia in size and shape. This review focuses on the implications of rhizobial dimorphism for symbiotic cooperation and conflict. For instance, indirect selection should be more important for evolution of symbiotic behavior in rhizobia descended from the undifferentiated clonemates of nonreproductive swollen bacteroids than for those descended directly from bacteroids.

We have previously shown that soybean (Glycine max), which hosts reproductive bacteroids, can impose fitness-reducing sanctions on rhizobia that fix less nitrogen. Do legumes that host nonreproductive bacteroids impose similar sanctions? First we examine how legume host sanctions maintain cooperation in the legume–rhizobia symbiosis. We then consider the following questions, some of which cannot yet be answered definitively. How often have host traits that suppress bacteroid reproduction been gained or lost over the course of legume evolution? Does the increase in size characteristic of nonreproductive bacteroids directly or indirectly benefit the plant? Do nonreproductive bacteroids have fewer opportunities to redirect resources from nitrogen fixation to rhizobial reproduction? If so, what are the implications for evolution of rhizobia and legumes? Are legume sanctions universal?

1. Sanctions against reproductive bacteroids maintain rhizobial mutualism

Symbiotic partners can have conflicting interests as well as shared interests. Successful symbioses have various mechanisms that can align the interests of the partners or enforce cooperation despite conflicting interests (Douglas, 2008). Two important mechanisms are (i) vertical transmission of symbionts from a host to its own offspring, and (ii) host sanctions.

Rhizobia spread to new hosts through the soil, not via seeds, and thus mutualism cannot be stabilized by vertical transmission. Even if they were vertically transmitted, the usual presence of multiple strains per individual plant would create a potential tragedy of the commons (Denison, 2000). The tragedy is that the individual benefit to a rhizobial strain of diverting resources from nitrogen fixation to its own reproduction outweighs the shared cost of reduced photosynthesis in a nitrogen-deficient host plant. Mathematical modeling has shown that, with a realistic number of rhizobial strains per individual host plant, strains that invest little or nothing in nitrogen fixation will outcompete those that invest more (West et al., 2002b).

Why do these cheaters, strains that potentially benefit from investing less in nitrogen fixation, not spread through the population? Experiments using nitrogen-free air (Ar:O₂) showed that soybean plants impose sanctions on individual nodules that fix less nitrogen (Kiers et al., 2003). Strain-dependent differences in nodule growth consistent with sanctions have also been reported for wild Lupinus arboreus (Simms et al., 2006).

Intermediate rates of nitrogen fixation lead to sanctions with intermediate effects on rhizobial fitness, but rhizobia whose nitrogen fixation is between 50 and 100% that of the best strains may pay little fitness penalty for cheating (Kiers et al., 2006). However, strains that fix even less than that still persist. Rhizobia isolated from the same soil can vary tenfold in plant benefit (Burdon et al., 1999). A few strains result in less legume growth than an uninoculated control (Nutman, 1954).

If sanctions are so effective, why do we find less beneficial strains in nature? Possible explanations for the persistence of these strains, despite sanctions, include mixed nodules, conflicting selection regimes, biochemical manipulation of legumes by some strains of rhizobia, and differences in sanctions among host genotypes (Kiers & Denison, 2008). The frequency of mixed nodules has rarely been measured under field conditions. Up to 32% of field soybean nodules contained two strains (Moawad & Schmidt, 1987), which might keep the total nitrogen fixation per nodule high enough to avoid sanctions, even if one strain fixed little nitrogen. There is also evidence that soybean cultivars differ in their yield response to mixtures of fixing and nonfixing rhizobia under field conditions, perhaps as a consequence of differences in sanction strength (Kiers et al., 2007). This could result in the escape of less beneficial strains. Rhizobia that block plant ethylene signaling are less beneficial to plants, but nonetheless acquire more resources per cell (Ratcliff & Denison, 2009), a possible example of manipulation. All of these experiments used hosts in which rhizobial bacteroids retain the ability to reproduce. Are rhizobial interactions with other legume host species similar?

2. Reproductive potential of bacteroids depends on host species

In pea (Pisum sativum) nodules, bacteroids are four to seven times the size of free-living rhizobia (Oke & Long, 1999) and can no longer reproduce (Kijne, 1975; Mergaert et al., 2006). However, these nodules also contain clonally identical undifferentiated rhizobia that do retain the ability to reproduce, but do not fix nitrogen. This is analogous to social insects with worker and reproductive castes. By contrast, in beans (Phaseolus vulgaris), bacteroids continue to divide, are similar in size to free-living cells, and are reproductively viable (Fig. 1b–d). The size distribution of the rhizobial cells from a single bean nodule is unimodal, while swollen bacteroids and undifferentiated rhizobia of pea nodules result in a clear bimodal distribution (Fig. 1c). We refer to bacteroids as swollen if there are distinctly smaller, undifferentiated rhizobia in the same nodule, resulting in rhizobial dimorphism, whether or not there are differences in shape.

Figure 1. The same Rhizobium strain is dimorphic in pea (Pisum sativum) (right) but not in bean (Phaseolus vulgaris) (left). Dimorphic rhizobia are found in many vicioid legumes with indeterminate nodules while legumes with determinate nodules, such as bean and Lotus, contain a homogenous bacteroid population (a). Bacteroid morphologies are evident using microscopy (b), including electron micrographs, and flow cytometry (c). Bacteroid viability (d) is often assessed by comparing total number of cells with number of colony-forming units (CFUs) on plates. A ratio significantly greater than 1 and nonviable cell counts equal to the swollen-cell population suggest that bacteroids are nonreproductive. When bacteroids are swollen, future hosts are infected mainly or exclusively by descendants of undifferentiated rhizobia (e). Polyhydroxybutyrate (PHB) is common in nonswollen bacteroids but usually absent from swollen ones (f) (Lodwig et al., 2005). Differences in PHB and reproductive viability suggest different cheating strategies for rhizobia in different host species. Nodule (birdsfoot trefoil and alfalfa) photos by Inga Spence.

The generalization that swollen bacteroids have little or no capacity to divide is based on evidence that varies among species. For example, Zhou et al. (1985) used video microscopy to show that only undifferentiated rhizobia divide in Trifolium repens nodules. Ratcliff et al. (2008) also showed a correlation between colony counts on growth agar and the number of smaller sized (undifferentiated) rhizobia in alfalfa (Medicago sativa) nodules via flow cytometry. More generally, swollen bacteroids are usually found individually within symbiosomes, compartments delimited by plant membranes, whereas a symbiosome may contain several nonswollen bacteroids (Izaguirre-Mayoral & Vivas, 1996; Denison, 2000). Swollen bacteroids tend to have altered cell wall structures that have higher osmotic sensitivity (Sutton & Paterson, 1979). Some studies suggest that swollen bacteroids can revert back to the rod-shaped reproductive forms (Gresshoff et al., 1977; Khetmalas & Bal, 2005). Although some swollen bacteroids may have this capability to dedifferentiate back to reproductive forms, current evidence suggests that the majority of viable rhizobia escaping into the soil are descended from the undifferentiated cell populations in the nodules, not from these swollen forms (Fig. 1e) (Zhou et al., 1985, Timmers et al., 2000; Ratcliff et al., 2008). Another character often attributed to swollen bacteroids that will not be explored in this review is genome endoreduplication, which is thoroughly examined by Maunoury et al. (2008).

Experiments with dual-host rhizobia show that rhizobial dimorphism depends on the legume host. Rhizobia whose bacteroids were swollen in vetch nodules (Vicia sativa) were genetically modified to infect Lotus japonicus (Bras et al., 2000), where they formed small, rod-shaped bacteroids like those typically found in that host (Mergaert et al., 2006). Conversely, rhizobia that formed small bacteroids in bean were modified to infect pea (Gotz et al., 1985), where they formed swollen bacteroids typical of that host (Mergaert et al., 2006). Another dual-host rhizobial strain (not genetically modified) makes swollen bacteroids in peanut (Arachis hypogaea) nodules and nonswollen ones in cowpea (Vigna unguiculata) nodules (Sen & Weaver, 1980). The direct mechanism by which bacteroids lose reproductive viability is unknown.

Lodwig et al. (2003) reported a mechanism for nutrient exchange between pea and its Rhizobium leguminosarum bacteroids, which superficially resembles sanctions. In pea nodules, bacteroids apparently depend on their host for essential amino acids. It was suggested that this dependence imposes ‘selective pressure for the evolution of mutualism’ (Lodwig et al., 2003). This hypothesis assumes that bacteroids with more access to amino acids will have greater fitness. If pea bacteroids could reproduce, the extent to which their reproduction depends on amino acids from the plant would have to be determined. As noted above, however, pea is one of the host species in which rhizobia are dimorphic, with the bacteroids having no descendants (Kijne, 1975). This is a fundamental difference from bacteroids in soybean nodules.

II. Evolutionary history: is host-imposed rhizobial dimorphism ancestral or derived in legumes?

Studies on bacteroid differentiation are limited to a few legumes, mostly species with agricultural benefits (Table 1). It remains to be seen how host-imposed rhizobial dimorphism is distributed in the legume phylogeny. From the current literature, it appears there may be at least two independent lineages of legumes whose nodules host both undifferentiated rhizobia and larger bacteroids: species within the vicioids, which include alfalfa and pea, and some aeschynomenoid-nodule-forming legumes such as peanut. However, a comprehensive phylogenetic analysis is required to determine if host suppression of bacteroid reproduction is derived or ancestral in the legume family.

Table 1.  Bacteroid properties and nodule types in representative legume species

	Alfalfa, pea, vetch	Soybean, bean, cowpea	Birdsfoot trefoil	Lupine	Peanut
References: 1, Banba et al. (2001); 2, Fernández-Pascual et al. (2007); 3, Gerson et al. (1978); 4, Mergaert et al. (2006); 5, Sen & Weaver (1984); 6, Sprent et al. (1987) and references therein; 7, Sprent (2001); 8, Sutton & Paterson (1980); 9, Ratcliff et al. (2008); 10, Trainer & Charles (2006) and references therein; 11, van Slooten et al. (1990); 12, Zhou et al. (1985).PHB, polyhydroxybutyrate.
Bacteroid size	5–10 µm long (4, 9, 12)	0.5–1 × 2–3 µm; 3.5 µm^3 (1, 4, 5, 6, 9, 12)	0.2–0.5 × 1.5–2 µm (1, 6)	1.2–1.5 µm long (2)	11–16 µm^3 (5, 6)
Bacteroid reproduction	Not viable (4, 8, 11)	Viable (4, 8, 9, 12)	Viable (4, 8)	Low viability (8)	Low viability (8)
Bacteroid PHB	None (9, 10)	Up to 50% dry weight (1, 6, 10, 11)	Rare (1, 6)	> 500 µg g^−1 of nodule is possible (3)	?
Nodule type (7)	Indeterminate	Determinate	Determinate	Genistoid (indeterminate)	Aeschynomenoid (determinate)

Elucidating what form of bacteroid differentiation is the derived trait in legumes can help formulate further hypotheses regarding possible benefits to legumes of this trait. If ancestral legumes suppressed bacteroid reproduction, then several lineages, including those containing bean and birdsfoot trefoil (Lotus corniculatus), subsequently lost this ability to modify bacteroids. Repeated loss of a trait would suggest that the trait reduces legume fitness. Alternatively, ancestral legume nodules could have hosted nonswollen reproductive bacteroids. This would mean that host-imposed rhizobial dimorphism has arisen at least twice, in vicioids and aeschynomenoids, suggesting that rhizobial dimorphism is somehow more beneficial to legumes. These inferences are according to analysis of the currently accepted legume phylogenies (Doyle et al., 1997; Wojciechowski et al., 2004).

1. No consistent relationship between nodule type and rhizobial dimorphism

Nodules can be classified into two general types – indeterminate and determinate. Indeterminate nodules have a persistent meristem and typically elongate during growth, while determinate nodules have a transient meristem and appear spherical. Nodule types can be further classified into subtypes, as discussed in more detail by Sprent (2001, 2008). The most frequently studied legume species (alfalfa, pea, birdsfoot trefoil, bean and soybean) show a correlation between nodule type and bacteroid reproduction: among these species, indeterminate nodules have swollen bacteroids and determinate nodules do not (Fig. 1a,b, Table 1). Although a general relationship between bacteroid viability and nodule type has been suggested (Denison, 2000; Lodwig et al., 2005), that apparent pattern does not stand up to closer analysis. Presently, there is evidence that peanut and related legume species, which have aeschynomenoid nodules (a special type of determinate nodule), also have swollen (coccoid) bacteroids (Fleischman & Kramer, 1998) and may have low reproductive viability (Sutton & Paterson, 1980). Also, there are many other legume species not commonly studied, for example Lupinus albus (Table 1), that are known to have indeterminate nodules but appear to have nonswollen rod-shaped bacteroids (Fernández-Pascual et al., 2007) and no rhizobial dimorphism. Therefore, although ancestral legumes are thought to have had indeterminate nodules (Sprent, 2007), we cannot assume that they hosted dimorphic rhizobia with nonreproductive bacteroids.

III. Are there immediate benefits to individual plants from rhizobial dimorphism?

Rhizobia descended from the undifferentiated clonemates of nonreproductive swollen bacteroids could evolve differently from those descended directly from bacteroids. This is because selection for greater or lesser nitrogen fixation by nonreproductive bacteroids must act via effects on the survival and reproduction of their undifferentiated clonemates.

This difference in rhizobial evolution is likely to have long-term consequences for the legume species the rhizobia nodulate and may indirectly benefit future legume hosts. However, natural selection is driven by immediate benefits to individual plants, not future consequences for the species as a whole. Typically, a plant's effects on rhizobial evolution will not preferentially benefit that individual plant, although there may be some exceptions. For instance, legumes with more than one round of nodulation (e.g. perennials) might benefit from the evolution of greater mutualism in rhizobia that are likely to reinfect their own roots. However, any benefits of improving mutualism in soil populations of rhizobia would be shared with nearby competitors. These plants may be kin, because of limited dispersal, but the same limited dispersal makes competition more local, decreasing the effects of kin selection (Griffin & West, 2002). Therefore, in most cases, it is unlikely that host suppression of bacteroid reproduction first evolved (or was lost) because of its effects on future rhizobial evolution. Rather, we need to consider the immediate fitness effects, for an individual legume plant, of host-imposed rhizobial dimorphism. How might a legume plant benefit from suppressing bacteroid reproduction?

Mergaert et al. (2006) suggest that limiting bacteroid reproduction might prevent bacteroids from becoming parasitic and infecting other plant tissues. However, it has been claimed that endophytic rhizobia in rice (Oryza sativa) leaves have physiological effects that have even been reported to be beneficial under some conditions (Chi et al., 2005).

Alternatively, loss of reproductive viability in bacteroids could be a consequence of bacteroid swelling that benefits hosts. Swollen bacteroids may be advantageous for several reasons. Plants may have an easier time retrieving nutrients from swollen bacteroids during nodule senescence (Mergaert et al., 2006; Maunoury et al., 2008) because they are easier to lyse. Swollen bacteroids might also have better symbiotic performance (Mergaert et al., 2006), for example higher nitrogen fixation per gram of carbon invested in nodule construction or per gram of carbon respired. Apparent differences in fixation efficiency have been detected using the same rhizobial strain in different hosts. In peanut, where bacteroids are swollen, nitrogenase activity per mg of bacteroid protein was five times higher than in cowpea, where bacteroids are not swollen (Sen & Weaver, 1984). Witty et al. (1983) also compared a single strain in these two hosts and found no difference in the nitrogen fixation per unit respiration. However, they used detached nodules, which they later showed can give inaccurate results (Minchin et al., 1986).

If swollen bacteroids prove to be more efficient, what mechanisms are most likely? Their lower surface:volume ratio might reduce the cost of maintaining ion gradients. This could reduce the maintenance component of bacteroid respiration, freeing resources for nitrogen fixation. Sen & Weaver (1984) presented evidence that differences in oxygen supply (perhaps linked to differences in nodule development rather than bacteroid swelling per se) may affect nitrogen fixation rate per gram of nodule, but they did not measure the ratio of nitrogen fixation to respiration. With only one bacteroid per symbiosome, as is typical for dimorphic rhizobia, there may be more direct contact between symbiosome membranes and bacteroid cell walls, which could perhaps improve resource transfer between host and bacteroid.

When rhizobia are dimorphic, the swollen bacteroids typically accumulate little or no polyhydroxybutyrate (PHB), a high-energy storage compound. This potentially frees resources for nitrogen fixation, as discussed below in detail. Natural selection would therefore favor plant traits that directly or indirectly inhibit PHB synthesis by bacteroids. Could loss of bacteroid reproductive viability and bacteroid swelling be side effects of hypothetical manipulation of rhizobial PHB metabolism by the plant? We include this possibility for completeness, but know of no published evidence that host plants can suppress PHB accumulation by bacteroids in their nodules or that this would lead to bacteroid swelling.

Our working hypothesis is that plant traits leading to rhizobial dimorphism (with swollen bacteroids), loss of bacteroid reproduction, and inability of bacteroids to accumulate PHB are newly derived in several legume lineages. We further hypothesize that only dimorphism and bacteroid swelling are imposed directly by these legume hosts, probably by interfering with bacterial cell wall growth (Sutton & Paterson, 1983). Individual plants somehow benefit immediately from bacteroid swelling (e.g. increased nitrogen fixation or easier retrieval of nutrients from senescing bacteroids). Loss of reproductive potential of bacteroids could then be a side effect of swelling. Lack of PHB synthesis in swollen bacteroids could either be another side effect of swelling or the result of subsequent rhizobial evolution, which is discussed in the next section.

IV. How does dimorphism affect rhizobial evolution?

Imagine a species of rhizobia whose legume host has recently evolved to suppress all reproduction by bacteroids. How might this affect evolution of this rhizobia?

If every rhizobial cell inside a nodule became a nonreproductive bacteroid, it is likely that the rhizobia would soon abandon symbiosis altogether. But with dimorphism not all rhizobia in nodules become bacteroids. Nonreproductive, swollen bacteroids typically share nodules with undifferentiated, reproductive clonemates. An alfalfa nodule, for example, can contain several million reproductive rhizobia, in addition to the nonreproductive bacteroids (Ratcliff et al., 2008). So infecting alfalfa plants is not a reproductive dead end for rhizobia. However, if hosts that impose dimorphism on rhizobia thereby reduce the number of descendants a nodule-founding rhizobial cell is likely to have, that might increase the risk:benefit ratio of attempting to nodulate (Denison & Kiers, 2004).

How might rhizobial traits evolve that affect the behavior of rhizobia once inside nodules? Rhizobial traits that reduce the fraction of cells becoming bacteroids might be favored by natural selection among rhizobia, if bacteroids leave no descendants. Traits that allow rhizobia to fix nitrogen in symbiosis, but without swelling or losing reproductive viability, might also be favored. Is it possible that some rhizobia with nonswollen bacteroids have successfully resisted this form of manipulation by their hosts? There are claims of dedifferentiation by bacteroids (Gresshoff et al., 1977, Khetmalas & Bal, 2005), but this may simply reflect undifferentiated rhizobia in nodule zones occupied mainly by bacteroids. It does seem, however, that natural selection among rhizobia should favor bacteroids regaining the ability to divide, if that is possible.

1. How can nonreproductive bacteroids enhance their inclusive fitness?

In a host where bacteroids irreversibly lose the ability to reproduce, their evolutionary impact on the next generation will depend on how their activities in symbiosis affect the survival and reproduction of their reproductive clonemates, perhaps especially those in the same nodule.

In contrast to reproductive bacteroids, nonreproductive bacteroids seem less likely to hoard resources like PHB to support their own long-term survival (Fig. 1g), because they will have no direct descendants. Instead, there are at least three ways in which nonreproductive bacteroids might have adapted to benefit their reproductive clonemates in the same nodule. First, nonreproductive bacteroids might divert resources (such as rhizopines; discussed in the next section) to be consumed by their reproductive clonemates. This activity may compromise nitrogen fixation. Secondly, bacteroids might reduce their own rate of resource consumption, thereby freeing more resources for their reproductive clonemates. A simple way to do this would be to reduce respiration in support of nitrogen fixation. Thirdly, bacteroids might elicit host responses that increase net resources to their reproductive clonemates. For example, if the influx of host photosynthate delivered to nodules (and to the reproductive clonemates) were directly proportional to the efflux of fixed nitrogen by bacteroids, then nonreproductive bacteroids might increase their inclusive fitness simply by fixing more nitrogen.

2. Can nonreproductive bacteroids cheat?

If nonreproductive bacteroids are able to divert resources to their reproductive clonemates, and thereby increase their inclusive fitness at the expense of host fitness, that would be considered a cheating strategy. How to do this? Some nonreproductive bacteroids are known to produce rhizopines, which are compounds synthesized by nonreproductive bacteroids within nodules and catabolized by the undifferentiated rhizobial cells (Murphy et al., 1995).

It has been suggested that rhizopines help promote rhizobial mutualism via kin selection by increasing the flux of root exudates to related, reproductively viable rhizobia in the rhizosphere (Olivieri & Frank, 1994; Simms & Bever, 1998). Such a mechanism relies on the assumption that the rhizobia receiving the benefits are closely related to the rhizobia in the root nodules, and that this relatedness arises through limited dispersal (Bever & Simms, 2000). However, this form of kin selection has stringent requirements for spatial genetic structures of the bacterial population outside the nodule (Simms & Bever, 1998). Given that spatial structure also undermines cooperation by making competition more local (West et al., 2001), within-nodule kin selection is likely to be the only form of selection strong enough to consistently promote cooperation.

For rhizopine diversion to be an effective cheating strategy, three conditions must be met: rhizopine synthesis must divert energy away from nitrogen fixation; rhizopines must be consumed by closely related rhizobia; and catabolizing rhizopines must increase the fitness of those rhizobia. The undifferentiated clonemates of bacteroids within the same nodule seem most likely to meet these criteria. Another cheating strategy that might operate similarly to rhizopines could be excess hydrogen production. Nitrogen fixation always releases some hydrogen gas as a byproduct and some reproductive rhizobia can consume hydrogen (Ruiz-Argueso et al., 1979). Bacteroids that increased hydrogen production from its baseline rate of c. 25% of nitrogenase activity (e.g. to 50%) might thereby benefit their reproductive clonemates at the expense of nitrogen fixation.

Even without resource diversion via rhizopines or hydrogen, carbon not consumed by nonreproductive bacteroids may be diverted to the undifferentiated rhizobia simply by diffusion. This could select for nonreproductive bacteroids that fix less nitrogen and thereby free more carbon for use by their reproductive clonemates. Nodules with ineffective (i.e. nonfixing) nonreproductive bacteroids contained higher levels of starch (Ronson et al., 1981), but whether excess carbon is available to reproductive cells in infection threads is unknown.

Although these cheating mechanisms are plausible, the extent to which nonreproductive bacteroids actually cheat in these ways is unknown. Rhizopine genes have been reported only in rhizobia whose bacteroids are nonreproductive in their usual hosts (Wexler et al., 1995), consistent with resource diversion to reproductive clonemates, but hydrogen consumption has also been reported in reproductive bacteroids, where it may actually increase nitrogen fixation (Albrecht et al., 1979). It also remains uncertain whether carbon that is unused by bacteroids benefits undifferentiated rhizobia enough to select for lower nitrogen fixation.

If these cheating options are not available, then a nonreproductive bacteroid pays little or no opportunity cost when it consumes carbon to power nitrogen fixation. This contrasts with reproductive bacteroids, where there is a clear trade-off between nitrogen fixation and hoarding resources to support their own reproduction (Hahn & Studer, 1986; Ratcliff et al., 2008), as discussed in the next section.

3. Reproductive bacteroids often hoard PHB at the expense of nitrogen fixation

Many bacteria synthesize high-energy lipid polymers such as PHB and accumulate them as reserves of energy. Some symbiotic rhizobia can store some of the carbon they receive from their host as PHB. It has been suggested that this PHB may help fuel further nitrogen fixation during times of low photosynthesis (Kretovich et al., 1977; Gerson et al., 1978; Bergersen et al., 1991), without any detailed consideration of how doing so would affect rhizobial fitness. Accumulated PHB can also support rhizobial survival and even reproduction in the absence of an external carbon source (Ratcliff et al., 2008), which presumably enhances fitness of reproductive rhizobia in soil between hosts. PHB accumulation seems to be most common in reproductive bacteroids, undifferentiated rhizobia sharing nodules with nonreproductive bacteroids, and free-living rhizobia. These are also the forms of rhizobia capable of reproduction. PHB has been seen in electron micrographs as white granules inside bacteroids (Figs 1f, 2a,c) and detected by lipid stains or gas chromatography in many nonswollen, reproductively viable bacteroids, but it has yet to be reported in significant amounts in swollen, nonreproductive bacteroids (Kim & Copeland, 1996).

Figure 2. Differences in polyhydroxybutyrate (PHB) accumulation resulting from host-strain interaction in nonswollen bacteroids are evident from electron micrographs (from Banba et al., 2001). (a) Rhizobium etli CE3 fixing nitrogen in Lotus japonicus; (b) Mesorhizobium loti fixing nitrogen in L. japonicus; (c) R. etli CE3 fixing nitrogen in Phaseolus vulgaris. Short arrows indicate PHB; arrowheads indicate peribacteroid membrane; long arrows indicate disintegrated membranes. b, bacteroid; hc, host cytoplasm. Bars: (a, b) 0.5 µm; (c) 1 µm.

Because PHB synthesis and nitrogen fixation can compete for the same sources of reducing agents, there is a trade-off between the two processes within reproductive bacteroids, as shown in bean and soybean nodules. A PHB-minus mutant fixes more nitrogen in symbiosis with bean (Cevallos et al., 1996). In soybean nodules, a nonfixing mutant accumulates more PHB than its fixing parent (Hahn & Studer, 1986). In both cases, bacteroids were reproductive, suggesting a conflict of interest between the host and symbiont and a possible mechanism of cheating (Fig. 1g).

4. How did rhizobial evolution affect bacteroid PHB hoarding?

We hypothesize that rhizobial strains with a long history of interacting mainly with legume hosts in which bacteroids are nonreproductive have not been selected to hoard PHB. Bacteroids gain no direct fitness benefit by hoarding PHB if they themselves cannot have descendants. Furthermore, hoarding PHB could hurt their reproductive clonemates in the same nodule in two ways. First, PHB synthesis by bacteroids might reduce the total pool of carbon available to their reproductive clonemates. Secondly, diverting resources from nitrogen fixation to PHB could trigger nodule-level sanctions that hurt their reproductive clonemates. Rhizobial evolution would then favor strains in which nonreproductive bacteroids forgo PHB accumulation in order to maximize nitrogen fixation and avoid sanctions against their reproductive clonemates. Reducing or eliminating PHB accumulation may be the easiest way for nonreproductive bacteroids to increase fixation without paying an opportunity cost.

However, the first generation of bacteroids that gained or lost the ability to reproduce would retain traits from their ancestors. For example, if rhizobia descended from reproductive bacteroids infected a newly evolved host that suppressed bacteroid reproduction, bacteroids might still hoard PHB. They might eventually abandon bacteroid PHB hoarding, but this would require at least a few generations of rhizobial evolution. Similarly, rhizobia that are reproductive in one host might retain bacteroid PHB hoarding in an alternate host where they are nonreproductive, unless that host somehow interferes with PHB synthesis. This could be the case, for example, for rhizobia whose bacteroids are reproductive in cowpea but not in peanut. Even reproductive bacteroids may lack PHB, however, depending on their overall carbon balance or other factors (Fig. 2b).

V. How might sanctions differ when bacteroids are nonreproductive?

Reproductive bacteroids pay a direct fitness opportunity cost when they respire carbon to support nitrogen fixation rather than using it to support their own immediate reproduction, or hoarding it as PHB to support future reproduction (Ratcliff et al., 2008). Neither of these options is available to nonreproductive bacteroids, so we might expect rhizobial cheaters to be less common when bacteroids are nonreproductive. If this is true, are host sanctions still needed in these legume species? Are sanctions still possible, if bacteroids will have no descendants anyway?

1. Should hosts with nonreproductive bacteroids invest in sanctions?

We define sanctions as some action by the host plant that reduces the relative fitness of less beneficial rhizobia based on their low rate or low efficiency of nitrogen fixation. Reduced nodulation, perhaps as a result of incorrect recognition signals, would be an example of partner choice rather than sanctions (Kiers & Denison, 2008). Also, sanctions are defined by their effects on rhizobial fitness, not host fitness. However, selection among legumes will only maintain sanctions if individual plants that impose sanctions thereby increase their fitness. This legume fitness benefit need not depend on how a given legume response affects rhizobial evolution. Note that rhizobial cheaters may mimic recognition signals of mutualists or retain signals from mutualistic ancestors.

We previously showed theoretically that individual plants would benefit from reducing resource consumption in nodules containing less mutualistic rhizobia (West et al., 2002a). From the plant's point of view, this is simply an efficient use of resources. From the point of view of rhizobia, however, this preferential resource allocation could function as a sanction, reducing the relative fitness of less mutualistic strains.

If nonreproductive bacteroids have few cheating options, then legumes hosting dimorphic rhizobia may encounter less beneficial rhizobia less often than legumes that host reproductive bacteroids. For example, < 13% of Sinorhizobium meliloti and Rhizobium leguminosarum strains tested had rhizopine genes (Wexler et al., 1995). However, given the other possible cheating options discussed above, it might be premature to conclude that cheating is rare when bacteroids are nonreproductive.

Furthermore, even if rhizobia with nonreproductive bacteroids are rarely cheaters, that is, strains that potentially benefit from fixing less nitrogen, legumes may still face considerable variation in symbiont quality. Variation in symbiont quality could occur if rhizobia that are less beneficial to particular hosts arise as a result of conflicting selection imposed by different host genotypes (Heath & Tiffin, 2007). There could also be significant numbers of mutants that retain the ability to nodulate but are defective in nitrogen fixation, without necessarily benefiting from fixing less nitrogen (Denison & Kiers, 2004). Given such variation, legumes would benefit from preferential allocation of resources to the most effective rhizobia (West et al., 2002a), whether or not less effective bacteroids are actually cheating.

2. Among-bacteroid versus among-nodule sanctions

If even those legumes that suppress bacteroid reproduction have encountered less beneficial rhizobia often enough to select for some host response, what sort of response should we expect? The response favored by natural selection among legumes is not necessarily the one that most favors the evolution of rhizobial mutualism.

If mixed-strain nodules are rare, a plant could increase its nitrogen return per carbon investment by preferentially allocating more carbon to the best performing bacteroids, regardless of whether this allocation is among bacteroids within a nodule or among nodules. If mixed nodules are common, then legumes would benefit more from among-bacteroid allocation than from among-nodule allocation (Denison, 2000). With among-nodule allocation, some resources could be wasted on less beneficial bacteroids within better performing nodules. This is true whether or not bacteroids are reproductive.

Reduced carbon allocation to less mutualistic rhizobia might be achieved by a host-imposed limitation on some other resource, such as amino acids (Lodwig et al., 2003) or oxygen (Kiers et al., 2003), limiting carbon use indirectly. We expect most legume species to have at least one mechanism for preferentially allocating resources to the subset of rhizobia in their nodules that provide the most nitrogen, relative to their carbon consumption.

We referred to ‘allocation’ above to leave open the question of whether a given legume response constitutes a ‘sanction’, that is, whether it imposes selection for greater rhizobial mutualism. Differences in allocation by legumes among nonreproductive bacteroids will have no direct effect on rhizobial evolution, although it is conceivable that there could be an indirect effect. For example, a nonreproductive bacteroid that receives more carbon because it fixes more nitrogen might then send more carbon to its reproductive clonemates via rhizopines. This assumes that making fewer rhizopines was not the reason for its superior nitrogen fixation.

We do not know whether among-bacteroid allocation is physiologically possible, but this could be tested by comparing fitness of marked effective vs ineffective isogenic strains in mixed nodules. Among-nodule allocation would provide some efficiency benefit to the plant even if some nodules were mixed. Differential allocation among nodules would presumably affect whichever rhizobia are reproductive in those nodules. Whether or not it is the bacteroids that are reproductive, differential allocation to better performing nodules would favor rhizobial mutualism most if mixed nodules were rare. As discussed above, natural selection among legumes would not usually depend on how their responses to differences among bacteroids or nodules affect rhizobial evolution.

3. Testing for sanctions by comparing rhizobial strains

When we ask whether a legume host imposes sanctions, we are asking whether they respond to differences among strains in actual symbiotic performance, for example nitrogen return on carbon investment, in ways that reduce the fitness of less beneficial strains. We know that legumes also respond to recognition signals during initial infection and perhaps later. In this paper, we are not considering the extent to which such responses to signals, as opposed to responses to actual performance, are adaptive for legumes.

There are two different types of evidence suggesting that legume species that suppress bacteroid reproduction may also impose sanctions on poorly performing strains. So far, however, each approach has been implemented in ways that limit our ability to draw firm conclusions about these species. One approach is to compare rhizobial strains, to see whether those that provide substantially less benefit to a given host also obtain less benefit from that host, as would be expected if that host imposes sanctions. A second approach is to manipulate a strain's nitrogen fixation rate by adjusting nitrogen concentration.

One problem with comparing strains is that differences in rhizobial fitness could be a result of some host-strain interaction other than performance-based sanctions. For example, a strain that fixes nitrogen very efficiently (i.e. nitrogen fixed per carbon respired) in a given host might also have signal molecules or surface antigens that limit nodulation or nodule growth in that host. A comparison with another strain that fixes nitrogen less efficiently but forms more nodules or reproduces better in nodules (as a result of signaling interactions, not performance-based sanctions) could incorrectly be interpreted as evidence against host sanctions. Comparing a larger number of strains would reduce the chance of drawing incorrect conclusions from idiosyncratic associations between signals and symbiotic performance.

Comparing benefits provided and obtained by different strains is also more difficult than generally recognized. It is easy to compare the growth of plants each inoculated with a single strain of rhizobia. However, growth with single-strain inoculation is an imperfect proxy for differences in mutualism among strains under field conditions, where each plant is infected by multiple strains. With single-strain inoculation, a poorly nodulating strain with high nitrogen-fixation efficiency might result in less plant growth than a less efficient strain that forms more nodules. Plant growth with single-strain inoculation would incorrectly identify this as a less beneficial strain. In the field, however, the poorly nodulating strain might only occupy 10% of the nodules on a plant, while providing 20% of the nitrogen. A better method might be to compare the growth of plants with two different strains in their nodules, in different proportions. The more beneficial strain is the one that increases plant growth as its nodule representation increases. A benefit:cost ratio would also be a better measurement to assess a strain's symbiotic qualities. For example, the ratio of nitrogen fixation rate to respiration rate of a single nodule or the ratio of host shoot biomass to total nodule mass would be a more accurate evaluation of the strain's symbiotic quality, independent of differences in nodulation efficiency. After all, in the real world, no one strain makes the majority of nodules on a single host.

Measuring differences among strains in rhizobial fitness benefits from symbiosis is also nontrivial. Nodule number from single-strain inoculation is not an appropriate proxy for rhizobial fitness. With single-strain inoculation, the number of nodules formed may depend on whether a strain can suppress the plant's autoregulation of nodule number, for example by producing rhizobitoxine, which blocks ethylene signaling (Sugawara et al., 2006). A rhizobitoxine producer may increase the total number of nodules but not its own proportional representation (Ratcliff & Denison, 2009). Competitiveness for nodulation in the field depends on survival under field conditions and potentially on hostile interactions among rhizobia, such as through production of bacteriocins (Schwinghamer & Brockwell, 1978), neither of which is measured in single-strain inoculation assays. However, the number of nodules per plant in single-strain inoculation can be used to estimate a plant benefit per nodule, somewhat alleviating the problem discussed in the previous paragraph.

Nodule size is widely measured and more promising as a proxy for rhizobial fitness. In soybean, which hosts reproductive bacteroids, much of the difference in rhizobia per nodule with sanctions was explained by nodule weight (Kiers et al., 2003). Similarly, Heath & Tiffin (2007) reported a positive correlation (r = 0.59) between reproductive rhizobia (determined by plate counts) and nodule length in Medicago truncatula. However, the relationship between nodule size and the number of reproductive rhizobia inside may not always be comparable among different strains. Therefore, even if nodules containing cheaters have lower average sizes, this does not guarantee that they have fewer reproductive cells per nodule. Given this uncertainty, reliance on nodule size as a proxy for rhizobial fitness seems unnecessarily risky. We suggest actually counting viable rhizobia per nodule, which can be done using agar plates with an appropriate growth medium. We also recommend measuring PHB per cell, using flow cytometry, because rhizobial cells can accumulate enough PHB to support doubling or tripling of their numbers (Ratcliff et al., 2008). In light of our concerns about single-strain inoculation and nodule size, we hesitate to draw firm conclusions from most of the published data, which are somewhat contradictory.

Miller & Sirois (1982) measured growth of four alfalfa cultivars with each of five rhizobial strains. They found a high correlation (r = 0.78) between average nodule weight (perhaps representative of rhizobial benefits) and average yield per plant with single-strain inoculation (perhaps representative of host benefits) (Fig. 3). For this data set, the correlation was similar if plant benefit from rhizobia was corrected for nodules per plant. This is the same sort of correlation seen in soybean (Kiers & Denison, 2008), for which we have direct experimental evidence for sanctions (Kiers et al., 2003). However, we cannot exclude the possibility that healthier alfalfa plants simply support larger nodules (in addition to or instead of more nodules per plant). This could be true even if alfalfa has no ability to selectively support the best nodules within a plant.

Figure 3. Effect of five Rhizobium meliloti strains on four alfalfa cultivars, based on data from Miller & Sirois (1982).

Preferential host support of the best performing nodules can be detected using mixed-inoculation experiments. For example, Singleton & Stockinger (1983) found that soybean nodules containing nitrogen-fixing rhizobia were 2.5 times as big as nodules on the same plant that contained an ineffective strain. Heath & Tiffin (2009) also used mixed inoculation, testing three rhizobial strains on four M. truncatula families. Nodule length varied fourfold, but was not correlated with estimates of genotype-specific host benefits from each strain. This was interpreted as evidence against sanctions. However, host benefits of each strain were measured in a previous experiment that used single-strain inoculation. With that approach, differences in total nitrogen fixation per plant among strains may be attributable to differences in nodulation, rather than differences in nitrogen fixation efficiency or rate per rhizobial cell.

Another important point is that a given set of strains may not include any whose nitrogen fixation rate is low enough to trigger sanctions, even in hosts that do impose sanctions on the worst strains. Therefore, to use comparisons among rhizobia to test whether a given legume genotype can impose sanctions, one should include a nodulating but nonfixing strain. Comparing this nonfixing control with an isogenic fixing strain could reduce the chance of drawing incorrect conclusions about sanctions because of idiosyncratic signaling interactions. Whether rhizobia that are ineffective enough to trigger sanctions are common (or rare, as suggested by Heath and Tiffin for their field locations) is a separate question. Plant alleles for sanctions that are never used would tend to disappear, however, unless they serve some other function.

Given uncertainties about using nodule size as a proxy for rhizobial fitness and plant growth with single-strain inoculation as a proxy for rhizobial mutualism, as discussed above, we consider the results of Miller & Sirois (1982) to be weak evidence for sanctions in alfalfa and the results of Heath & Tiffin (2009) to be weak evidence against sanctions in M. truncatula. It seems likely that the apparent discrepancy between these two studies is a consequence of deficiencies in methods, rather than a fundamental difference between these related legume species.

These same methodological concerns also apply to studies of two other legume species that host dimorphic rhizobia. Nutman (1946) reported that some rhizobia that are ineffective on clover nonetheless produce large nodules. More recently, Laguerre et al. (2007) found rhizobia that produce very large nodules on pea, but result in less plant growth than other strains. Neither report determined whether these large nodules actually contained large numbers of reproductive rhizobia.

4. Manipulating nitrogen fixation to test for sanctions

Given the problems discussed in the previous section, we favor testing for sanctions by manipulating nitrogen fixation rate rather than comparing rhizobial genotypes that may differ in various ways. We do this by reducing the nitrogen concentration around one or more nodules, thereby limiting nitrogen fixation. Our approach also allows us to vary nitrogen fixation quantitatively and over time.

In soybean, which hosts reproductive bacteroids, nonfixing nodules (in Ar:O₂) grew less and contained fewer viable rhizobia per nodule (Kiers et al., 2003). Intermediate nitrogen concentrations, allowing intermediate rates of nitrogen fixation, led to intermediate sanctions (Kiers et al., 2006). We have used both single-nodule chambers and split-root chambers, with similar results.

We are currently conducting similar experiments using alfalfa and also plan to test other species hosting nonreproductive bacteroids. Meanwhile, there is some indirect evidence that host sanctions may be widespread. Minchin et al. (1983) found a physiologically similar response to decreasing nitrogen fixation in a variety of legume species, including some that host nonreproductive bacteroids. Blocking nitrogen fixation in nodulated roots (using either an nitrogen-free Ar:O₂ atmosphere or acetylene, a competitive inhibitor of nitrogenase activity) triggered a decrease in nodule respiration rate, subsequently linked to a decrease in nodule O₂ permeability. Although the fitness effects of decreased respiration for dimorphic rhizobia have not been measured, the decreased nodule O₂ permeability is the same response we saw in soybean (Kiers et al., 2003). It therefore seems plausible, although far from certain, that sanctions may operate in a wide variety of legume species, by similar mechanisms.

VI. Conclusions

It is evident from reviewing the literature that one of the largest gaps in our understanding of legume–rhizobia symbiosis is a widespread neglect of the implications of a fundamental difference in rhizobial life histories: bacteroids swell and lose reproductive viability in some legume hosts while in others they remain similar to free-living forms and continue to reproduce. This difference in life history of symbiotic rhizobia should have significant evolutionary consequences for the mutualism. We have outlined several key questions and hypotheses concerning this fundamental dichotomy in host effects on rhizobial life history and drawn some tentative conclusions.

First, it appears that suppression of bacteroid reproduction by legumes has arisen or been abandoned at least twice over the course of legume evolution. We hypothesize that rhizobial dimorphism and bacteroid swelling are caused by one or more derived traits of the host not found in ancestral legumes, but this is not yet certain.

Secondly, the evolutionary transition to (or possibly from) host-induced rhizobial dimorphism was probably driven by some immediate benefit to individual legume plants, not by subsequent evolutionary changes in rhizobia. For example, swollen bacteroids may fix nitrogen more efficiently or may lyse more easily during nodule senescence, facilitating recovery of nutrients by the host.

Thirdly, swollen bacteroids are typically nonreproductive, so diverting resources from nitrogen fixation to their own reproduction is not an option. This reduction in the opportunity cost of nitrogen fixation may select against PHB hoarding by bacteroids, perhaps freeing resources for additional nitrogen fixation. However, even nonreproductive bacteroids may cheat by diverting resources from nitrogen fixation to their reproductive clonemates in the same nodule, for example via rhizopines.

Finally, we predict that all legumes will have mechanisms to preferentially allocate resources to nodules containing strains that fix nitrogen more efficiently. This response would enhance legume fitness even if some less efficient strains were simply defective or incompatible, rather than cheaters that potentially benefit from fixing less nitrogen. It remains to be seen whether these plant responses meet our definition of sanctions by actually reducing the fitness of less beneficial rhizobia.

There are several related areas that would benefit from additional research. First of all, more information is needed on wild legumes. It is important to expand our knowledge beyond the typical agricultural host species and possibly look for exceptions and unique symbioses. Generalizations inferred from agricultural species should not be frivolously extended to wild legumes. For example, rhizobial dimorphism and bacteroid viability need to be assessed directly, not inferred from nodule type (determinate or indeterminate). The frequency of mixed nodules could also differ among legume species, leading to differences in their symbiotic rhizobial evolution.

Other areas requiring further investigation are roles of rhizopines in rhizobial and host fitness and the role of oxygen permeability in host sanctions. Classification of less beneficial and more beneficial strains could also be more carefully assessed, moving beyond plant benefit in single-strain inoculation. Lastly, experiments comparing host effects on strains with different symbiotic qualities should use methods of manipulation (e.g. Ar:O₂) or near-isogenic strains to avoid complications based on arbitrary strain signals. To determine rhizobial fitness, we also need to count reproductive rhizobia in fixing versus nonfixing nodules, not just measure nodule size or weight.

Punishment in social animals (West et al., 2007) or in cleaning mutualisms (Bshary et al., 2008) can increase cooperation by changing the behavior of individuals, but we have no evidence of similar changes in the behavior of individual rhizobial cells in response to legume sanctions. We assume that sanctions improve rhizobial behavior through evolution, but it would be worth checking for phenotypic plasticity in the behavior of individual rhizobia in response to sanctions.

Many details of host physiological responses to less beneficial strains and their evolutionary consequences for rhizobia remain to be elucidated. We expect these details to reveal similarities between symbioses with reproductive and nonreproductive rhizobial bacteroids, but also important differences.

Acknowledgements

ETK was supported by an NWO ‘Veni’ grant and RO and RFD by NSF grant 0514464.