1. Evolutionary Biology
Download icon

Spatio-temporal control of mutualism in legumes helps spread symbiotic nitrogen fixation

  1. Benoit Daubech
  2. Philippe Remigi
  3. Ginaini Doin de Moura
  4. Marta Marchetti
  5. Cécile Pouzet
  6. Marie-Christine Auriac
  7. Chaitanya S Gokhale
  8. Catherine Masson-Boivin  Is a corresponding author
  9. Delphine Capela
  1. Université de Toulouse, INRA, CNRS, France
  2. Massey University, New Zealand
  3. Fédération de Recherches Agrobiosciences, Interactions et Biodiversité, Plateforme d’Imagerie TRI, CNRS - UPS, France
  4. Max Planck Institute for Evolutionary Biology, Germany
Research Article
  • Cited 2
  • Views 897
  • Annotations
Cite as: eLife 2017;6:e28683 doi: 10.7554/eLife.28683

Abstract

Mutualism is of fundamental importance in ecosystems. Which factors help to keep the relationship mutually beneficial and evolutionarily successful is a central question. We addressed this issue for one of the most significant mutualistic interactions on Earth, which associates plants of the leguminosae family and hundreds of nitrogen (N2)-fixing bacterial species. Here we analyze the spatio-temporal dynamics of fixers and non-fixers along the symbiotic process in the Cupriavidus taiwanensis–Mimosa pudica system. N2-fixing symbionts progressively outcompete isogenic non-fixers within root nodules, where N2-fixation occurs, even when they share the same nodule. Numerical simulations, supported by experimental validation, predict that rare fixers will invade a population dominated by non-fixing bacteria during serial nodulation cycles with a probability that is function of initial inoculum, plant population size and nodulation cycle length. Our findings provide insights into the selective forces and ecological factors that may have driven the spread of the N2-fixation mutualistic trait.

https://doi.org/10.7554/eLife.28683.001

eLife digest

Rhizobia are soil bacteria that are able to form a symbiotic relationship with legumes – plants that include peas, beans and lentils. The bacteria move into cells in the roots of the plant and cause new organs called nodules to form. Inside the nodules the bacteria multiply before being released to the soil again. Also while in the nodules, the bacteria receive carbon-containing compounds from the plant. In return many of the bacteria convert (or “fix”) nitrogen from the air into compounds that the plant can use to build molecules such as DNA and proteins. Yet, some of the bacteria are “non-fixers” that provide little or no benefit to the host plant.

Evidence suggests that legumes select against non-fixer bacteria, though it was not clear when or how this selection process occurs. Daubech, Remigi et al. have now followed the number and viability of two variants of a bacteria species called Cupriavidus taiwanensis as they form a symbiotic interaction with Mimosa pudica, a member of the pea family. The two types of bacteria differed only by whether or not they were able to fix nitrogen. At first fixers and non-fixers entered nodules and multiplied at equal rates. Later, the fixers progressively outcompeted the non-fixers. Then, around 20 days after the bacteria entered the plant, nodule cells that contained non-fixers degenerated. This indicates that the nodule cells help to control bacterial proliferation based on the benefits they receive in return.

Further experiments and mathematical modeling also showed that over repeated cycles of root nodule formation, nitrogen fixers can invade a bacterial population dominated by non-fixer bacteria. The likelihood that this invasion will be successful increases as three other factors increase: the proportion of fixer bacteria in the initial population, the number of available plants, and the length of time the bacteria spend in the nodules. This mechanism ensures the maintenance and spread of nitrogen-fixing traits in the bacterial population.

Improving the processes of biological nitrogen fixation could help to reduce the amount of fertilizers required to grow crops. This in the future could help make agricultural ecosystems more sustainable. The results presented by Daubech, Remigi et al. provide guidelines that could be used to select nitrogen-fixing bacteria on legume crops or on nitrogen-fixing cereals that may be engineered in the future. Further work is now needed to understand in more detail the molecular mechanisms that lead to the death of non-fixer bacteria.

https://doi.org/10.7554/eLife.28683.002

Introduction

The evolutionary dynamics of mutualistic interactions between higher organisms and microbes depends to a large extent on the transmission mode of microbial symbionts. Vertical transmission is expected to promote fitness alignment of obligate symbionts and their partners (Herre et al., 1999). In contrast, horizontal transmission generates more complex ecological cycles for facultative symbionts. When going through these cycles, microbes are subjected to several trade-offs regarding host range (specialist vs. generalist) and investment in the mutualism (good or bad cooperator, life in the host vs. outside the host). The large number of possible strategies to maximize fitness, and the ability to segregate in a population of genetically variable partners, often entails conflicts of interests between symbionts and their hosts (Bever et al., 2009; Sachs et al., 2010; Porter and Simms, 2014; Jones et al., 2015) that may result in the classic Tragedy of the Commons (Hardin, 1968). The emergence and stability of mutualism thus requires that proliferation of symbionts is allowed but restricted to appropriate spaces and times and that beneficial partners are ultimately favored over uncooperative ones (Vigneron et al., 2014; Visick and McFall-Ngai, 2000; Koch et al., 2014). The theoretical aspects of the evolution and maintenance of mutualistic interactions have been extensively discussed (Archetti et al., 2011; Akcay, 2015). Yet experimental assessment is scarce and the impact of ecological factors, such as population size of hosts and symbionts or the duration of the interaction, has been under-explored, although they are an essential component of the evolutionary potential of symbiotic systems.

Rhizobia, the N2-fixing symbionts of legumes, induce the formation of and massively colonize nodules, where intracellular bacteria fix atmospheric nitrogen for the benefit of the plant in exchange for photosynthates. When the nodule senesces, nodule bacteria are released to the soil where they can return to free-living lifestyle and/or colonize a new host (Thies et al., 1995). During evolution, symbiosis modules carrying genes essential for the symbiotic process have spread to many different taxa so that extant rhizobia are distributed in hundreds of species in 14 genera of α- and β-proteobacteria (Remigi et al., 2016). Acquisition of symbiotic genes may not be sufficient to create an effective symbiont and may lead to bacteria exhibiting various levels of symbiotic capacities (Nandasena et al., 2006; Nandasena et al., 2007; Marchetti et al., 2010) that can be further optimized and maintained under legume selection pressure (Marchetti et al., 2017; Marchetti et al., 2014). It has been established that bacteria better able to form and infect nodules are selected by a partner choice mechanism involving the specific recognition of bacterial molecular signals by plant receptors (Kawaharada et al., 2015; Radutoiu et al., 2003). Bacterial features that are recognized by the plant include Nod factors that initiate rhizobial entry and nodule formation (Perret et al., 2000; Broghammer et al., 2012), and lipo/exopolysaccharides critical for root infection and bacterial release inside the plant cell (Kawaharada et al., 2015), as well as an array of bacterial effectors that refine host specificity (Deakin and Broughton, 2009). Nitrogen fixation however is uncoupled from nodulation and infection, and legumes can be nodulated and infected by ineffective symbiotic partners (Gehlot et al., 2013; Gourion et al., 2015). The emergence of mutualism in populations resulting from the transfer of symbiosis modules, and its maintenance over evolutionary timescales (Werner et al., 2014) indicates that the cooperative behaviour of the bacterial symbionts is controlled at the infection and/or post-infection levels by one or a combination of mechanisms. Partner choice is the selection of appropriate symbionts at the (pre-) infection stage based on signal recognition while post-infection sanctions rely on the ability to discriminate between low- and high-quality cooperators during an established interaction and to punish or reward them accordingly (Kiers and Denison, 2008; Frederickson, 2013). Partner-fidelity feedback (PFF) ensures positive assortment of symbionts during long lasting or repeated interactions in spatially structured environments independently from any recognition process or conditional response (Sachs et al., 2004). These different control mechanisms have been proposed to affect the dynamics of mutualistic traits, particularly in the context of the nitrogen-fixing symbiosis (Kiers et al., 2003; Oono et al., 2009). Here we evaluate how selective forces and ecological factors act on the dissemination of the nitrogen fixation mutualistic trait on the Cupriavidus taiwanensis-Mimosa pudica mutualistic interaction. Specifically we evaluated the spatio-temporal dynamics of N2-fixing and non-fixing bacterial subpopulations to model the spread of the N2-fixation trait across plant generations.

Results

Evidence for a spatial and temporal control of mutualism in Mimosa nodules

During the symbiotic process, most rhizobia enter the legume root via infection threads that ensure colonization of the forming nodule and ultimately release bacteria into nodule cells where differentiated forms called bacteroids fix nitrogen (Batut et al., 2004). Although they induce the formation of indeterminate nodules, it is noteworthy that Cupriavidus taiwanensis symbionts of Mimosa spp. are not terminally differentiated and ca. 20% of bacteroids recovered from nodules, together with bacteria present in infection threads, can resume growth (Marchetti et al., 2011). To evaluate the specific fates of mutualists and non mutualists in plants infected by a mixed population, we monitored the fitness of total nodule bacteria over time following co-inoculation of Mimosa pudica seedlings with a mixture (1/1 ratio, 106 total bacteria/plant) of isogenic N2-fixing and non-fixing strains of C. taiwanensis. Fix+ and Fix- strains only differed by the presence of the nifH gene, encoding the nitrogenase reductase subunit of the nitrogenase enzyme, and of constitutively expressed GFP or mCherry fluorescent genes. For technical reasons (see Materials and methods), nodules were only collected from 14 dpi. Importantly, each nodule was individually analyzed for bacterial fitness by plating, allowing analysis at the nodule and plant individual levels. In these experimental conditions 97% of the nodules were infected by either Fix+ or Fix- bacteria.

We observed a marked difference in the reproductive fitness of Fix+ and Fix- bacteria from the same plant over time, which significantly differed from 21 days post-infection (dpi) and up to 28 fold on average (Figure 1A and Figure 1—figure supplement 1), perhaps because of plant control mechanisms, including sanctions (Kiers et al., 2003) and possibly PFF. A significant difference was also obtained from 28 dpi when analyzing control plants singly-infected with either Fix+ or Fix- strains (Figure 2A). Non-fixers did not proliferate better than fixers even at 14 dpi (Figure 1A) possibly because the metabolic cost paid by bacteria to fix nitrogen in terms of ATP and reducing power is too low to be detected in our experimental conditions, or because plant sanctions/PFF and the metabolic cost of nitrogen fixation equilibrate until sanctions become prominent. The resulting net fitness cost of cooperation, which is the weighted metabolic cost of nitrogen fixation by any form of plant control, thus appeared to be zero or negative, enabling mutualism to spread.

Figure 1 with 1 supplement see all
Kinetics of reproductive fitness of Fix+ and Fix- bacteria in nodules following co-inoculation of M. pudica.

M. pudica plants were co-inoculated with a mixture of Fix+ and Fix- strains at a 1/1 ratio, using 106 (A) or 1010 bacteria/plant (B). Nodules were individually analyzed by plating their bacterial population (see Figure 1—figure supplement 1). Co-infected nodules represented ca. 3% (A) or 20% (B) of the nodules. (A) The ratio of the mean number of bacteria per Fix+-containing nodule to the mean number of bacteria per Fix-- containing nodule was calculated for each individual plant at each time point (see Figure 1—figure supplement 1) and box plots represent the distribution of these ratios (Figure 1—source data 1). Only single-infected nodules were taken into account in this graph. (B) Box plots represent the distribution of the ratios of Fix+ bacteria to Fix- bacteria in co-infected nodules (Figure 1—source data 2). Central rectangles span the first quartile to the third quartile (that is, the interquartile range or IQR), bold segments inside rectangles show the median, unfilled circles indicate suspected outliers, whiskers above and below the box show either the locations of the minimum and maximum in the absence of suspected outlying data or 1.5 × IQR if an outlier is present. Horizontal dashed lines correspond to ratios equal to 1. The number of plants (A) or nodules (B) analyzed is indicated in brackets. *Significant differences between the number of Fix+ and Fix- bacteria per nodule (p<0.05, multiple comparison test after Kruskal-Wallis (A); p<0.001, after Student t-test with paired data (B).

https://doi.org/10.7554/eLife.28683.003
Kinetics of reproductive fitness of Fix+ or Fix- nodule bacteria following single-inoculation of M. pudica.

(A) Fix+ (CBM382) or Fix- (CBM2568) C. taiwanensis were inoculated on M. pudica. Box plots represent the distribution of the number of bacteria recovered per nodule on plates. Box plots were constructed as described in Figure 1. R, ratios of the median number of Fix+ bacteria per nodule on the median number of Fix- bacteria per nodule. The number of nodules analyzed at each time point is indicated in brackets. The number of plants analyzed at each time point is indicated in red. Results are from two independent experiments (Figure 2—source data 1). *Significantly different from the number of Fix+ bacteria per nodule (p<0.05 multiple comparison test after Kruskal-Wallis). (BC) Theoretical reproductive fitness of Fix+ (B) and Fix- bacteria (C) following single-inoculation of M. pudica as compared to experimental data. Dotted lines represent bacterial populations per plant averaged over 200 replicate simulations (Figure 2—source data 2). Box plots represent the distribution of the number of bacteria experimentally recovered per plant. Experimental data are from (A).

https://doi.org/10.7554/eLife.28683.007

The differential fitness was not due to a better nodulation competitveness of Fix+ bacteria. The number of nodules formed by each strain was indeed proportional to the inoculum ratio (1/1) throughout the time course (Figure 3), confirming that bacterial nitrogen-fixing ability is not selected at the root entry level (Hahn and Studer, 1986; Westhoek et al., 2017). Yet the number of nodules in nitrogen-starved non-fixing plants (infected with 99% or 100% Fix-) constantly increased over a 42 day period, while this number reached a plateau at ca. 20 dpi in healthy N2-fixing plants (infected with 50% or 100% Fix+) (Figure 4), indicative of a mechanism of autoregulation of nodulation acting at the whole-plant level (Ferguson et al., 2010) and depending on the nitrogen status of the plant (Malik et al., 1987; van Noorden et al., 2016). This difference in time course increases the chance that a rare Fix+ among a Fix- population will form a nodule .

Relative number of nodules formed by Fix+ and Fix- bacteria per plant individual.

M. pudica plants were co-inoculated with the CBM2700 (Fix+, GFP) and CBM2707 (Fix-, mCherry) strains at a 1/1 ratio. The number of plants analyzed for each time point is indicated in brackets. Boxplots were constructed as described in Figure 1. No significant differences were observed between the number of nodules formed by Fix+ bacteria and Fix- bacteria per plant at the different time points (p>0.05, Student t-test with paired data at each time point or multiple comparison test after Kruskal-Wallis on the whole dataset) (Figure 3—source data 1).

https://doi.org/10.7554/eLife.28683.010
Nodulation kinetics.

M. pudica plants were single-inoculated with either CBM832 (Fix+) or CBM2568 (Fix-) or co-inoculated with a mixture of both strains at a 1/1 or 1/99 ratio. First nodules appeared at 5–7 dpi (Figure 4—source data 1).

https://doi.org/10.7554/eLife.28683.012

To identify the spatial level at which selection applies we first analyzed double occupancy nodules, which were obtained in significant proportion by modifying the plant culture system and increasing the inoculum density by four logs (see Materials and methods). Co-infected nodules contained a similar number of Fix+ and Fix- bacteria at 14 dpi, but on average ca. 80 times more N2-fixing bacteria than non-fixing bacteria at 35 dpi (Figure 1B), indicating that the control occurs at the nodule scale. Previous studies established that bacteroids do not persist in nodule cells of nitrogen-starved plants infected only by non-fixers, leading to premature nodule senescence (Berrabah et al., 2015; Hirsch and Smith, 1987), while they persist in healthy plants singly-infected with fixers. We therefore then analyzed the viability of bacteroids on sections of singly-occupied or double-occupied nodules collected from co-inoculation experiments using propidium iodide (PI), which stains dead cells (Virta et al., 1998). Bacteroid viability in Fix+-occupied nodules remained stable from 14 to 35 dpi (Figure 5D). By contrast, bacteroids in the nitrogen-fixing zone of Fix--occupied nodules started losing viability at 16–21 dpi and were all dead (PI-stained) at 35 dpi (Figure 5E). Electron microscopy confirmed signs of nodule cell and bacterial degeneration in Fix--occupied nodules at 19 dpi (Figure 6). Co-infected nodules showed clear sectoring, with infected plant cells in one part filled with Fix+ strains and in the other part filled with Fix- strains (Figure 5FGHI). We never observed co-infected nodule cells. While at 14 dpi both strains were alive (Figure 5G), at 35 dpi only Fix- bacteroids were PI-stained confirming that Fix+ and Fix- intracellular bacteria have distinct fates within the same nodule (Figure 5HI). The ca. 5 × 106 bacteria recovered at 35 dpi from nodules infected with only Fix- bacteria may thus be bacteria colonizing the infection threads and the infection zone.

Viability of Fix+ and Fix- bacteroids.

M. pudica were co-inoculated with Fix+ and Fix- C. taiwanensis at a 1/1 ratio and sections of nodules collected at 14 dpi (ABFG), 16 dpi (C) or 35 dpi (DEHI) were observed under bright field (panels 1) or fluorescent microscopy (panels 2 and 3), and after PI staining (panels with an *). Panels with the same letters represent the same nodule section. (F3), magnification of (F2) visualized by confocal microscopy. (A) and (D), sections of nodules infected with a GFP-labeled Fix+ strain. (B) (C) and (E), sections of nodule infected with a GFP-labeled Fix- strain. (F), nodule co-infected with a GFP-labeled Fix+ and a mCherry-labeled Fix- strain. (G) and (H), nodules co-infected with a GFP-labeled Fix- and an unlabeled Fix+ strain. (I), nodules co-infected with a GFP-labeled Fix+ and an unlabeled Fix- strain. The white and yellow dotted lines in (GHI) delimit the areas occupied by the Fix- and Fix+ strains in a co-infected nodule, respectively. Note that neither the Fix+ (D3) nor the Fix- bacteroids (B3G3) are red-labeled by PI staining at 14 dpi whereas a few cells are PI-stained in the Fix--occupied nodule at 16 dpi ([C3], arrows), and Fix- are mostly PI-labeled (dead) at 35 dpi (E3H3I3). Note that bacteria of the infection zone are still alive at 35 dpi (arrow, E2E3). Note that nodule cells filled with Fix- are browner than nodule cells filled with Fix+ (G1H1I1). Scale bars correspond to 100 µm except for F3 (30 µm).

https://doi.org/10.7554/eLife.28683.014
Electron microscopy of Fix+- and Fix--occupied nodules.

M. pudica plants were co-inoculated with Fix+ (CBM2708, mCherry) and Fix- (CBM2568, unlabeled) C. taiwanensis at a 1/1 ratio. Nodules collected at 19 dpi (ABCDE) were sorted for mCherry expression under fluorescence microscopy and used for electron microscopy observation. Degenerated nodule cells (*) were observed in Fix--occupied nodules (BDE) but not in Fix+-occupied nodules (AC). (C) and (D) represent magnification of the zones delimitated by a black dashed rectangle in (A) and (B) respectively. (E) magnification of the white rectangle in (D) showing degenerated bacteria (arrows). Scale bars represent 20 µm (ABC), 10 µm (D) and 2 µm (E).

https://doi.org/10.7554/eLife.28683.015

In conclusion we provide evidence for differential spatio-temporal dynamics of N2-fixing and non-fixing partners during the symbiotic process, highlighting the importance of considering temporal variations when studying the evolution of cooperative interactions (Barker and Bronstein, 2016). We established that the control of mutualism (i) acts at the nodule cell scale, (ii) occurs relatively early, ca. 16–21 days after inoculation when the wild-type nitrogenase is fully active in Fix+ bacteria (Figure 7) and (iii) leads to up a ca. 80 fold relative increase in mutualistic partners.

Kinetics of nitrogenase activity in N2-fixing M. pudica nodules.

Plants were inoculated with C. taiwanensis CBM832 (Fix+), and nitrogenase activity measured using the acetylene reduction assay (ARA) (Figure 7—source data 1). Two negative controls, i.e. tubes containing only the acetylene substrate and plants inoculated with C. taiwanensis CBM2568 (Fix-), were included. In these cases, boxplots correspond to data from all time points. *, Significantly different from the negative controls (p<0.05 after multiple comparison test of Kruskal-Wallis).

https://doi.org/10.7554/eLife.28683.016

Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles

Next, we addressed the question of whether mutualism control will allow a minority Fix+ subpopulation to invade the symbiotic population.

We first used our experimental data to develop a stochastic mathematical model qualitatively simulating the fate of C. taiwanensis populations during nodulation in M. pudica plants. The two key components of this model are (i) the kinetics of nodule formation from bacteria randomly chosen from the rhizospheric population and (ii) bacterial multiplication within nodules, according to bacterial genotype (see Materials and methods and Table 1 for details on model construction and parameterization). While the model is developed as a proof-of-concept, instead of a simple deterministic model we chose to include stochasticity in the nodulation process in order to reflect the variability observed in the experimental data. In order to test our model, we first simulated the reproductive fitness of nodule bacteria following single-inoculation with either Fix- or Fix+ bacteria over a 49 day-period, and compared this simulation to the kinetics experimentally observed (Figure 2BC). We then both simulated and experimentally determined the relative proportion of Fix+ bacteria recovered from plants co-inoculated with a minor subpopulation of Fix+ (1%) and a major subpopulation of Fix- (99%) bacteria over 49 days (Figure 8). Simulation outcomes qualitatively matched the dynamics of bacterial populations observed experimentally (Figures 2BC and 8), indicating that the experimentally measured and inferred model parameters are appropriate for studying the evolutionary dynamics of C. taiwanensis populations in different ecological conditions.

Experimental and theoretical reproductive fitness of Fix+ and Fix- bacteria following co- inoculation of M. pudica (ratio 1/100).

The proportion of Fix+ clones in nodules was experimentally measured and simulated over 49 days, following co-inoculation of 20 plants. Experimental data are shown as black triangles (Figure 8—source data 1). Black error bars represent standard deviation from 2 to 3 replicates. The results from 100 replicate simulations are shown as grey dots and boxplots (Figure 8—source data 2).

https://doi.org/10.7554/eLife.28683.018
Table 1
Model parameters
https://doi.org/10.7554/eLife.28683.021
ParameterAbbreviationValue
Size of each pool of plants*PoolVariable (1–1000)
Number of replicates*RepVariable (5 or 100)
Length of each cycle*DaysVariable (14-49)
Number of cycles*CycVariable (4 or 10)
Initial proportion of Fix+ cells*xVariable (1 or 0.1)
Maximum number of new nodules/plant/dayλmax0.44
Coefficient for the auto-regulation of nodulation in nodulation kineticsa10.03
Coefficient for time-decay in nodulation kineticsa20.006
Lag for time-decay in nodulation kineticsa32
Growth rate of bacteria within noduler1.95
Fitness cost of nitrogen fixation c0
Sanctions for Fix-‡s1.65
Day at which additional sanctions beginds17
Nodule carrying capacityK1.4 × 108
  1. *parameters varied in the simulations

    experimentally measured parameters

  2. parameters inferred from experimental data

We then used this model to explore how plant population size and the length of inoculation cycles impact on the dynamics of C. taiwanensis populations during serial cycles of inoculation of M. pudica plants and re-isolation of bacteria from nodules. Starting with a fixed proportion of Fix+ bacteria (1% or 0.1%) in the inoculum, we varied the number of inoculated plants from 1 to 100 (or 1 to 1000) and the length of nodulation cycles (time from plant inoculation to nodule bacteria harvesting) from 14 to 49 days, which is shorter than the lifespan of a nodule in nature. We found that larger plant pools and longer cycles progressively reduced extinction probabilities and increased the proportion of Fix+ in the nodule bacterial population (Figure 9A and Figure 9—figure supplement 1). For example, the model predicted that using an initial inoculum of 1% Fix+, 4 cycles of 42 days with pools of 20 plants were sufficient to yield more than 85% of Fix+ bacteria in all replicates where Fix+ populations avoided extinction (89 times out of 100 replicates in Figure 9A). Smaller plant pools or shorter cycles all yielded higher probabilities of extinction and decreased proportions of Fix+ bacteria. An initially lower Fix+ proportion (0.1%) could be compensated for by a higher plant population size and/or a longer cycle length (Figure 9—figure supplement 1). We analyzed in detail the dynamics of Fix+ subpopulations over 10 cycles in a situation where the cycle length had a major impact on the evolutionary outcome (20 plants) (Figure 9A) and plotted the proportion of Fix+ bacteria recovered after each cycle, for cycles ranging from 14 to 49 days (Figure 9B). We observed that, in the vast majority of cases, the fate of Fix+ populations is already determined after the first cycle: these populations are either bound to extinction (with a probability indicated in Figure 9A) or to a gradual increase in frequency that ultimately leads to fixation. This result holds true for all cycle lengths except 14 days, where population dynamics is dominated by drift due to the equivalent fitness of Fix- and Fix+ clones (Figure 1A). A key factor controlling the early bifurcation between extinction and fixation of Fix+ population is the probability that a Fix+ bacterium forms a nodule during the first cycle, which depends on both the size of plant pools and the length of nodulation cycles.

Figure 9 with 1 supplement see all
Effect of cycle length and plant numbers on the predicted distributions of Fix+population sizes.

Model simulations were performed with an initial proportion of 1% Fix+ in the bacterial population inoculated to a pool of plants. The length of each cycle and the number of plants per pool varied as indicated in the legend. (A) Final proportion of Fix+ clones after four cycles (Figure 9—source data 1). Boxplots represent the distribution of the final proportion of Fix+ clones from 100 simulations. The length of inoculation cycles ranged from 14 to 49 days and the number of plants per pool from 1 to 100. Numbers underneath each boxplot indicate the number of replicate simulations where Fix+ sub-populations became extinct after four cycles. (B) Increase in the proportion of Fix+ clones along 10 inoculation cycles of 14, 21, 28, 35, 42 or 49 days (Figure 9—source data 2). The number of plants per pool was 20. Representative trajectories of 5 replicate pools are shown in each case.

https://doi.org/10.7554/eLife.28683.022

Understanding the influence of plant pool size is straightforward. Very few nodules are produced on each plant, creating a bottleneck in bacterial population size at each nodulation cycle. Whatever the cycle length, larger numbers of plants per pool increase the likelihood that at least one Fix+ clone is sampled from the rhizospheric population, giving Fix+ subpopulations an opportunity to increase in frequency and avoid extinction in the next cycle (Figure 9A and Figure 9—figure supplement 1). Under longer cycles, extinction probability decreases (Figure 9AB) since more nodules are produced (Figure 4) and the size of Fix+ populations increases at a faster rate (Figure 9B) as a result of a decrease in Fix- fitness in older nodules (Figure 1A). The combined action of these two factors act on the inoculum for next cycle, generating an eco-evolutionary feedback.

To assess the predictions of the model experimentally, we performed serial inoculation-nodulation cycles of 21 or 35 days using 20 M. pudica plants and an initial inoculum of 5 × 103 Fix+/5 × 105 Fix- C. taiwanensis per plant. In each 35 day-cycle the nitrogen-fixing subpopulation increased and it reached nearly 100% of the population after four cycles (Figure 10A), similar to what observed with the model. Under 21 day-cycles, both simulations and experiments lead to a slower progression of Fix+ subpopulations (Figure 10B). It is worth noting that an increase in frequency of the best cooperators among natural strains was also observed after three consecutive nodulation cycles between Medicago truncatula and Sinorhizobium meliloti (Heath and Tiffin, 2009), indicating that the selective advantage of the best N2-fixing strains seems to be robust to the natural diversity of symbiotic associations.

Frequency of Fix+ bacteria over 4 cycles of 35 (A) or 21 (B) days: simulations and experimental validation.

The proportion of Fix+ clones over four inoculation cycles was simulated and measured experimentally. Simulations and experiment were performed with an initial proportion of Fix+ clones of 1% and pools of 20 plants. Experiments were performed with an inoculum of 5 × 103 Fix+/5 × 105 Fix- C. taiwanensis per plant. The results from 100 replicate simulations are shown as grey dots and boxplots (Figure 10—source data 1). Experimental data are shown as black triangles (Figure 10—source data 2).

https://doi.org/10.7554/eLife.28683.027

Discussion

Identifying the selective forces and ecological factors that shape mutualism is central to predicting its maintenance and dissemination over evolutionary scales. Here we provide conclusive evidence that nitrogen fixation per se, the ultimate trait that turns a parasitic rhizobium-legume association into a mutualistic one, determines the in planta spatio-temporal fate of endosymbiotic bacteria. Non-N2-fixing symbionts do not persist within cells of indeterminate nodules even when they share a nodule with N2-fixing symbionts, indicative of a cell autonomous senescence program as recently shown for determinate nodules (Regus et al., 2017). This results in the progressive and selective in planta expansion of fixers during the symbiotic process.

The most likely explanation is that the plant exerts a post-infection control of N2-fixation that overcomes the metabolic cost of nitrogen fixation paid by mutualistic bacteria. Sanctions could occur as defense responses and/or by decreasing nutrient supply to non-fixing bacteroids. Given that Fix- and Fix+ bacteria are spatially segregated within nodules, the latter case could also result from the local degeneration of nodule cells, and be interpreted as an example of Partner Fidelity-Feedback mechanism occurring at the level of individual cells (Shou, 2015). Since control mechanisms prevent social dilemma –i.e. the possibility that one partner increases its own fitness by decreasing its investment in mutualism- and help cooperation persist (Kiers and Denison, 2008; Frederickson, 2013; Sachs et al., 2004), non-fixers do not threaten mutualism in our system. Yet the fate of strains able to fix intermediate levels of nitrogen fixation may be different. Monitoring the fitness of strains varying in their nitrogen fixation capacity would provide a more complete picture of mutualism control. Nevertheless, our results provide an additional example supporting the emerging idea that low quality rhizobial partners rarely benefit from low investment in mutualism (Jones et al., 2015; Friesen, 2012). Plant sanctions resulting in bacterial fitness reduction were demonstrated in some rhizobium-legume systems by simulating N2 deficiency via gas manipulation around nodules (Kiers et al., 2003; Oono et al., 2011), although not seen in other systems (Marco et al., 2009; Ling et al., 2013). That different plants may rely on different control mechanisms would not be surprising given the variety of mechanisms that lead to symbiosis with legumes (Masson-Boivin et al., 2009).

Experimental investigations can fuel a theoretical framework able to reframe general evolutionary questions in an ecological context (Hoek et al., 2016). Our qualitative model of the eco-evolutionary dynamics of mutualistic and non-mutualistic populations includes serial inoculation-nodulation cycles. This regime mimics an experimental set up of horizontal transmission of rhizobia across plant generations albeit on an accelerated basis. A general outcome of the model is that rare fixers will invade a population dominated by non-fixing bacteria, above a threshold combination of plant and bacterial population sizes and cycle lengths. The model helps explore further combinations of number of cycles, cycle lengths and plant pool sizes to hypothesize the evolutionary trajectory of the Fix+ genotype. While the selective advantage of the Fix+ phenotype is expected to ensure its fixation in a deterministic manner, strong population bottlenecks occurring at the nodulation step introduce a source of stochasticity in these dynamics and may thus prevent the action of directional selection. The effect of stochasticity has been shown to be of immense evolutionary consequence in related models of host parasite coevolution (Papkou et al., 2016). Another characteristic of our system is that, when the Fix+ populations increase in abundance then so does their proliferation, leading to a quick increase of Fix+ over successive nodulation cycles (Figure 9B). This interaction between the demographic composition of the population and the evolutionary success of one of the traits is an example of the eco-evolutionary feedback present in this system.

Although the selective and ecological forces at play in the lab and in field conditions may differ significantly, our results predict that both forces have played a major role in the evolution of the rhizobium-legume mutualism by favoring the fixation of emerging N2-fixing sub-populations among uncooperative symbiotic populations as well as their evolutionary maintenance. Yet the uncooperative population does not become extinct within nodules, likely because sanctions mainly target bacteroids of the nitrogen fixation zone. Releasing non-fixing bacteria may allow progenitors to meet appropriate hosts or to evolve new symbiotic traits. This loose selection process helps maintain genetically diverse rhizobial communities in the soil and shape the ecology and evolution of rhizobia. More generally, acknowledging the existence of non-cooperators as an integral component of the ecological and evolutionary dynamics of mutualistic interactions may provide a better understanding of the long-term persistence of bacterial lineages (Heath and Tiffin, 2009; Heath and Stinchcombe, 2014; Tarnita, 2017; Fiegna et al., 2006; Hammerschmidt et al., 2014).

An emerging trend in fundamental and applied plant microbiology is to select upon microbes indirectly through the host (Mueller and Sachs, 2015). This engineering approach, called host-mediated selection, involves selection of microbial traits that are not selectable in vitro. Modelling the eco-evolutionary scenarios provides predictions to guide experimental evolution studies aiming at designing beneficial microbes (Marchetti et al., 2010; Marchetti et al., 2017) and microbiomes (Mueller and Sachs, 2015; Johns et al., 2016).

Materials and methods

Bacterial strains and growth conditions

Strains and plasmids used in this study are listed in Table 2.

Table 2
Strains and plasmids used in this study
https://doi.org/10.7554/eLife.28683.030
BacteriumStrainRelevant characteristicsReference/source
E. coliDH5αF recA lacZM15Bethesda research laboratory
DH5α λpirF recA lacZM15 λpirHP Schweizer
C. taiwanensisLMG19424Wild-type strain isolated from Mimosa pudica in Taiwan(Chen et al., 2001)
CBM832LMG19424 derivative resistant to Streptomycin, StrRM. Hynes
CBM2568CBM832 deleted in nifH, StrRThis study
CBM2700CBM832 carrying a Pps-GFP fusion downstream glmS, StrRThis study
CBM2701CBM2568 carrying a Pps-GFP fusion downstream glmS, StrRThis study
CBM2707CBM2568 carrying a Pps-mCherry fusion downstream glmS, StrRThis study
CBM2708CBM832 carrying a Pps-mCherry fusion downstream glmS, StrRThis study
PlasmidsNameRelevant characteristicsReference/source
pGPI-SceIoriR6K, mob+, carries a I-SceI site, TriR(Flannagan et al., 2008)
pDAI-SceIoripBBR1, mob+, carries the I-SceI gene, TetR(Flannagan et al., 2008)
pRCK-Pps-GFPPlasmid carrying the psbA promoter region fused to GFP, KanRM. Valls
pRCK-Pps-mCherryPlasmid carrying the psbA promoter region fused to mCherry, KanRM. Valls
pCBM156pGPI-SceI carrying the nifH 5’ and 3’ regions, TriRThis study
pCBM161pGPI-SceI carrying the glmS-Ralta_A0206 intergenic region interrupted by a Pps-GFP fusion, TriRThis study
pCBM162pGPI-SceI carrying the glmS-Ralta_A0206 intergenic region interrupted by a Pps-mCherry fusion, TriRThis study
pRK2013Helper plasmid, KanR(Figurski and Helinski, 1979)
  1. Str, spreptomycin; Tri, trimethoprim; Tet, tetracycline; Kan, kanamycin.

C. taiwanensis strains were grown at 28°C on TY medium supplemented with 6 mM CaCl2 and 200 µg/ml streptomycin. E. coli strains were grown at 37°C on LB medium and antibiotics were used at the following concentrations: kanamycin 25 µg/ml, trimethoprim 100 µg/ml, tetracycline 10 µg/ml. For in vitro competition experiments, strains were pre-cultured in TY medium, mixed in equal proportion then co-inoculated to a 100 ml culture in TY medium. Bacteria were plated every 2 hr during the exponential phase, at the entry of stationary phase and 15 hr after the entry into the stationary phase. Plated bacteria were grown for 48 hr at 28°C then green and red bacteria were counted using a fluorescence stereo zoom microscope (Axiozoom V16, Zeiss).

Mutant construction

Mutant and labeled strains of C. taiwanensis were constructed using the mutagenesis system developed by Flannagan et al. (Flannagan et al., 2008) involving the suicide plasmid pGPI-SceI carrying an I-SceI recognition site and the pDAI-SceI replicative plasmid expressing the I-SceI nuclease. To construct the unmarked C. taiwanensis nifH mutant, regions upstream and downstream nifH were amplified with the oCBM1821-oCBM2362 and oCBM1822-oCBM2363 primer pairs using GoTaq DNA polymerase (Promega). The two PCR products were digested with XbaI-BamHI and BamHI-EcoRI respectively and cloned into the pGPI-SceI plasmid digested by XbaI and EcoRI. Ligation products were transformed into a DH5α λpir E. coli strain. The resulting plasmid was transferred into C. taiwanensis CBM832 by triparental mating using pRK2013 as helper plasmid. Transconjugants that have integrated the plasmid by single crossing over were selected on streptomycin and trimethoprim and verified by PCR using the oCBM1824-oCBM2363 and oCBM1825-oCBM2362 primer pairs. Then we introduced the pDAI-SceI replicative plasmid into these strains by conjugation and selection on tetracyclin. Expression of the I-SceI nuclease causes a double strand break into the inserted plasmid and promotes DNA recombination. Mutants deleted in nifH were screened by trimethoprim sensitivity and verified by PCR using the oCBM1824-oCBM1825 pair of primers. Mutants were then cultivated on unselective TY medium. Tetracycline sensitive colonies which have lost the pDAI-SceI plasmid were selected.

The Pps-GFP and Pps-mCherry fusions were inserted into the wild-type and nifH mutant of C. taiwanensis at the same chromosomal locus, i. e. in the intergenic region between the glmS and RALTA_A0206 genes using the same pGPI-SceI/pDAI-SceI mutagenesis system. Flanking regions of the insertion site were amplified by PCR using the Phusion DNA polymerase (ThermoFisher Scientific) and the oCBM2619-oCBM2620 and oCBM2621-oCBM2622 primer pairs. PCR products were digested by XbaI and Acc65I or Acc65I and EcoRI respectively and cloned into the pGPI-SceI plasmid digested by XbaI and EcoRI. The two fusions Pps-GFP and Pps-mCherry were obtained by digesting the pRCK-Pps-GFP and pRCK-Pps-mCherry by AvrII and SpeI and cloned into the pGPI-SceI carrying the intergenic region glmS-RALTA_A0206 digested by the same enzymes. The resulting pCBM161 and pCBM162 were first transformed into a DH5α λpir E. coli strain then transferred into C. taiwanensis by triparental mating with the pRK2013 helper plasmid. Integration of the fusions by double crossing over was carried out using the pDAI-SceI plasmid as described above. CBM2700 (Fix+, GFP) and CBM2707 (Fix-, mCherry) had the same plating efficiency in in vitro competition experiments, indicating that these genetic modifications did not noticeably affect bacterial growth rate.

Oligonucleotide sequences used for genetic constructions are provided in Supplementary file 1.

Plant tests

Mimosa pudica seeds were of Australian origin (B and T World Seed, Paguignan, France) and were sterilized as described (Chen et al., 2003). Seedlings were cultivated in Gibson tubes (2 M. pudica plantlets/tube) as previously described (Marchetti et al., 2014). To increase the frequency of co-infection, plants were grown on 12 cm2 plates (three plants per plate) containing slanting nitrogen-free Fahraeus agar medium for 3 days at 28°C. Roots were covered with a sterile, gas-permeable, and transparent plastic film (BioFolie 25; Sartorius AG, Vivascience, Bedminster, NJ, U.S.A.). For single-strain inoculation experiments, each plant in Gibson tubes was inoculated with 5.105 bacteria either CBM832 (wild-type) or its isogenic nifH mutant, CBM2568. For co-inoculation experiments in Gibson tubes, plants were inoculated with the two isogenic strains CBM2700 (wild-type, GFP labeled) and CBM2707 (nifH, mCherry labeled) at ratio 1/1 (5.105 bacteria of each strain per plant) or 1/100 (5.103 bacteria of CBM2700 and 5.105 bacteria of CBM2707 per plant). For co-inoculation experiments in plates, plants were inoculated with 1010 bacteria of each strain per plant.

To measure the number of nodule bacteria over time, all nodules from 5 to 10 individual plants, except very small nodules, were individually collected with at least 2 mm of root left on both sides of nodules and treated at each time point. We did not collect very small nodules since there was a risk that the sterilization agents penetrate these nodules. In the same line we did not collect nodules before 14 dpi since most nodules were very small at that stage. Nodules were surface sterilized for 15 min in a 2.5% sodium hypochlorite solution, rinsed with water and crushed. Each nodule crush was diluted and plated using an easy spiral automatic plater (Interscience). Colonies were counted after 2 day-incubation at 28°C, under a fluorescence stereo zoom microscope (Axiozoom V16, Zeiss) when appropriate.

For nodulation kinetics, the number of nodules formed on 20 plants grown in Gibson tubes was counted daily for 6 weeks.

For serial inoculation-nodulation cycles on M. pudica plants, 10 Gibson tubes of plants were inoculated with CBM2700 and CBM2707 in 1/100 ratio as described above. 35 days after inoculation, all nodules were collected, surface-sterilized and crushed together. The nodule crush was used to inoculate a new set of 10 tubes of plants with 50 µl of a 1/10 dilution of the nodule crush per plant. At each cycle, dilutions of the nodule crush were spread on plates, incubated 2 days at 28°C and colonies were counted under a fluorescence stereo zoom microscope.

Cytological analyses

The viability of nodule bacteria was estimated using propidium iodide staining at a concentration of 20 mM in DMSO (Molecular Probes, Fisher scientific, Oregon) on 55/58 µm nodule sections. For each experiment, a dozen nodules were individually analyzed at 14, 16, 17, 21, 28 and 35 dpi. For electron microscopy analysis, nodules were fixed in glutaraldehyde (2.5% in phosphate buffer 0.1 M [pH 7.4]), osmium treated, dehydrated in an alcohol series, and embedded in Epon 812. Semithin nodule sections were observed by brightfield microscopy after staining in 0.1% aqueous toluidine blue solution and observed under a Zeiss Axiophot light microscope. Ultrathin sections were stained with uranyl acetate and observed with a TEM Hitachi HT7700.

Acetylene reduction assays

M. pudica plants were inoculated with the wild-type strain of C. taiwanensis CBM832. At different time points, plants were removed from the culture Gibson tube and placed in an airtight tube and incubated with 1 ml of acetylene for 4 hr. 100 µl of gas were then injected into a gas chromatograph (Agilent GC7820). The area of the ethylene peak was measured and compared to an ethylene standard of known concentration. Ethylene background was estimated by analyzing empty tubes incubated with the same amount of acetylene.

Mathematical model and simulations

The model aimed at simulating nodulation dynamics during single or repeated inoculation-nodulation cycles. First we parameterized the population dynamics during the symbiosis process. Then we simulated repeated nodulation cycles varying the following parameters: (i) the Fix+/Fix- ratio in the initial inoculum, (ii) the number of inoculated plants, and (iii) the cycle length. The model ran on a pool of plants (of given, variable size) from which nodules were collected and mixed together after each inoculation cycle. For each time-step (1 day) after inoculation, the number of new nodules formed on each plant was randomly drawn from a Poisson distribution of parameter λ(t, nod+t), which is itself a function of time t and of the number of nodules already present on the plant nod+t at time t. The maximal number of nodules that could potentially be formed per day per plant was set to λmax. Changing the value of parameter λ depending on the number of Fix+nodules already present on the plant simulated the autoregulation of nodulation process; this was done by subtracting the factor a1 × nod+t from λmax. Lastly, to allow for some ‘aging’ process that would decrease the rate of nodulation with time (even for plants inoculated only with Fix- bacteria), we incorporated a time-decay coefficient: a2 × (t- a3), meaning that a reduction in the rate of nodulation occurred at a rate a2 when t > a3. This time-decay factor was set to 0 when t < a3. Therefore, the parameter of the Poisson distribution controlling the rate at which new nodules are formed was given by: λ(t, nod+t) = λmax - a1 × nod+t for t < a3 and by: λ(t, nod+t) = λmax - a1 × nod+t - a2 × (t- a3) for t > a3. Since nodules are persistent once formed, we further set: λ(t, nod+t)≥0. Experimental evidence indicated that the number of inoculated bacteria did not affect nodulation kinetics as long as the total inoculum remains above 103 bacteria per plant. These conditions were met in all experiments described in this work. Therefore, we did not explicitly take inoculum size into consideration in the simulations, and restricted the applicability of our model to cases where inoculum was above this threshold value.

The second module of the model dealt with bacterial multiplication within plant nodules. Within each nodule we assumed a logistic growth model for the bacteria given by: X(t + 1) = (r-c –suds(t))×X(t) × (1-X(t)/K), where r was the growth rate, c the net fitness cost of nitrogen fixation in Fix+ bacteria, suds(t) the additional plant sanctions against Fix- bacteria occurring in the later phase of the interaction, X(t) the bacterial population at time t and K the nodule carrying capacity. In our simulations, we set c = 0 since we experimentally did not detect any difference in the populations of Fix- or Fix+ nodule bacteria at 14 dpi. We emphasize that a net fitness cost of 0 does not necessarily imply that nitrogen fixation does not impose a metabolic burden on the bacteria (referred to as ‘metabolic cost’ in the results section). Instead, this burden, if significant during the early steps of the interaction, may be compensated for by plant control mechanisms acting at a basal level. Beyond this time point, additional plant sanctions (possibly including partner fidelity-feedback) were given by suds(t), taking the value s of plant sanctions indicated in Table 1 as long as the age of the nodule was higher than ds days (denoted by the step function uds(t)=0 if t < ds or uds(t)=1 if t > ds).

Parameters values were estimated by computing the minimal root mean square error (RMSE) of experimental data (nodulation kinetics and bacterial multiplication within nodules) versus model outputs calculated for a range of parameter values. Parameter values selected to minimize RMSE are indicated in Table 1. Simulations were implemented in R (R Core Team, 2014) and code is available in the Source code file 1.

References

  1. 1
    Evolutionary models of mutualism
    1. E Akcay
    (2015)
    In: J. L Bronstein, editors. Mutualism. New York:  Oxford University Press. pp. 57–76.
    https://doi.org/10.1093/acprof:oso/9780199675654.003.0004
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
    R: A language and environment for statistical computing
    1. R Core Team
    (2014)
    R Foundation for Statistical Computing, Vienna, Austria.
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
    Determination of complement-mediated killing of bacteria by viability staining and bioluminescence
    1. M Virta
    2. S Lineri
    3. P Kankaanpää
    4. M Karp
    5. K Peltonen
    6. J Nuutila
    7. EM Lilius
    (1998)
    Applied and Environmental Microbiology 64:515–519.
  61. 61
  62. 62
  63. 63

Decision letter

  1. Wenying Shou
    Reviewing Editor; Fred Hutchinson Cancer Research Center, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Spatio-temporal control of mutualism in legumes helps spread symbiotic nitrogen fixation" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom, Wenying Shou, is a member of our Board of Reviewing Editors. Ian Baldwin oversaw the process as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Maren Friesen (Reviewer #2) and R. Ford Denison (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

Baubech et al. examined how legumes deal with rhizobium that does not fix nitrogen. We find parts of the paper very interesting. However, we have problems with your data interpretation, especially when you regard nitrogen fixation as cost-free in your model and in your text. Seems that your results are consistent with partner-fidelity feedback control of non-fixing bacteria (i.e. host cells with fix- mutants don't do well, and thus fix- mutant do not do well). We encourage you to rethink and rewrite.

Reviewer #1:

My feelings for this paper are mixed. I like how the authors showed that nitrogen-starved plants had increased nodules over time (presumably until plant's nitrogen need is met). In a single nodule, fix+ and fix- nodule cells are spatially partitioned, and fix+ increases its frequency over fix-. Fix+ can increase in frequency during several rounds of inoculation/growth, and this increase is faster in longer cycle (as predicted by theoretical work decades ago). However, the paper made quite problematic assertions, which rendered the model and the entire paper problematic.

Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”: "Since non-fixers do not proliferate better than fixers (Figure 1A), they are not cheaters in our system following a fitness-based definition of the term "cheater". This statement (and the thinking behind it, including assuming that nitrogen-fixation has zero cost in modeling) is problematic. You are dealing with a spatially-structured environment. Given the spatial segregation of fix+ and fix- cells to different nodule cells, partner fidelity feedback (see Bull and other's work, which is further extended in https://elifesciences.org/articles/10106) cannot be ruled out. In fact, partner fidelity feedback is likely to operate. In other words, plant cells near fix- bacteria do poorly, and thus fix- bacteria do poorly as well. If you stand behind your claim that N2-fixation is cost free, then you must back up this extraordinary claim with exceptionally careful experiments. That is, if you were to mix the two types together so that they are well-mixed in a plant-like environment (which is probably difficult to do and not done here), fix- cells are more fit or less fit than fix+ cells?

Figure 1A, Figure 3: I understand fix+ and fix- bacteria, but I am confused about fix+ and fix- nodule. What is the definition?

Figure 2B/C: I am not convinced: there are no data points before dpi 14, and so growth rate cannot be credibly estimated.

Reviewer #2:

This is a significant contribution to the field of rhizobia-legume symbiosis, demonstrating nicely that non-fixing rhizobia are penalized both at the whole nodule level as well as within nodules – a matter of much debate! – and going farther to considering the time-course of this process. The modeling is a nice addition, though not necessarily required for this to be a strong paper. The multigenerational experiments seal the deal, demonstrating that the phenomenon measured within a generation translates to the expected shifts in allele frequencies over time.

I have only two concerns:

First, as this is something I worry about a lot in my own work, to what extent does the surface sterilization process penetrate and kill rhizobia within Mimosa nodules? This could have dramatic effects on the rhizobial populations observed in single-strain nodules that vary in size, which typically correlates with fixation status.

Second, the authors claim in the text that the model fits their experimental data very well, but do not justify this statistically. In fact, looking at Figure 2 and Figure 10, the model doesn't appear to fit all that well for particular timepoints and there doesn't seem to be all that much power in rejecting the model, if this was in fact the intent. Some consideration of what the purpose of the model is would be warranted – I think it is a nice conceptual addition that shows qualitatively that cycle length and plant population size impact the evolution of fix+ from low frequency, but that the quantitative fit is being overstated.

Reviewer #3:

This paper presents some important data, although some of the interpretation is questionable. The important results are:

1) Host sanctions against a nonfixing rhizobial strain were much more severe than previously reported for other host species (Figure 1A). In the real world, however, nonfixing nodulating strains like that used in the experiments are rare, relative to less-beneficial strains that fix some nitrogen. Furthermore, there was no evidence that the resources the nifH strain spared by not fixing nitrogen are automatically invested in rhizobial fitness. Therefore, we cannot conclude (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”) that "low quality rhizobial partners rarely benefit from low investment in mutualism." (If there were a consensus on this, as claimed, this paper would merit publication only if it challenged that consensus.) To test that hypothesis would require comparing two nitrogen-fixing strains, with one of them providing less benefit via a mechanism that would plausibly enhance its own fitness. Examples include strains that (once inside nodules) block nodulation by other rhizobia (Tatsukami, 2016) or strains that divert more resources from N2 fixation to their own reproduction. Nonfixing mutants would not qualify.

2) Sanctions were apparently imposed on a nonfixing strain within mixed nodules (Figure 1B). This is arguably the most-important result, so should be featured in the Abstract, along with the "clear sectoring" (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules") that is presumably key to within-nodule discrimination by the host.

3) Figure 3 shows no partner choice (host-imposed discrimination during the infection process). This is consistent with past work comparing isogenic strains, but the authors should recognize the possibility of adaptive partner choice by some host plants among some real-world strains whose signals are correlated with their relative benefits.

Given these empirical results, it's obvious that fix+ rhizobia could invade a fix- population under real-world conditions. I don't really see the point to modeling cases with only 20 plants or where plants only live 14 days after rhizobial infection, as either of those would lead to extinction of the host. If the authors want to explore "ecological factors", as claimed, they should explore some more-realistic factors like temperature or soil nitrogen. What about competition between a strain that fixes half as much nitrogen as another, diverting the resources saved to its own reproduction? That's more realistic than a nonfixing strain. Their relative fitness would depend on how the host's threshold for imposing sanctions and on the extent to which resources diverted from nitrogen fixation actually enhance rhizobial fitness. Until we have those data, modeling is almost pointless. If space in the journal is not an issue, they could keep the modeling work, but the Abstract should be rewritten to be less vague and to highlight the empirical results. One sentence in the Abstract on modeling would be plenty.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for submitting your revised article "Spatio-temporal control of mutualism in legumes helps spread symbiotic nitrogen fixation" for consideration by eLife. Your revision has been evaluated by Ian Baldwin in consultation with a Reviewing Editor.

Thank you for your revision, but we are returning the revision without further review, because we don't think that you have adequately addressed the concerns about the "costs" that play a central role in your modeling effort (or we were not sufficiently clear in our previous decision letter). In your model, the "cost of nitrogen fixation" being zero strains credulity unless you are very specific about what that these costs mean.

Hence, we think that you need to be very specific about how these costs are described; otherwise the modelling effort could be criticized as being a tautology. If we understand your arguments correctly, the "post-PFF/post-sanctioning costs" need to be very small or even negative (benefit of cooperation) for cooperation to persist. Hence the text should be revised to say something like "the cost of cooperation is difficult to determine because such determination requires a well-mixed environment to be free of PFF. Regardless, this cost must have been overcome by PFF and/or sanctioning mechanisms. The cost we refer to is post-PFF/post-sanctioning cost of cooperation, which should be zero or negative, otherwise the mutualisms would have failed".

We hope that you can address this concern in short order and we look forward to subjecting your revision to a full re-review.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for submitting your article "Spatio-temporal control of mutualism in legumes helps spread symbiotic nitrogen fixation" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom, Wenying Shou is a member of our Board of Reviewing Editors and the evaluation has been overseen by Ian Baldwin as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Maren Friesen (Reviewer #2); R. Ford Denison (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

Reviewers feel that you have made improvements to the manuscript. There are points that still need to be addressed.

Essential revisions:

1) Since their most interesting result was apparent sanctions within mixed nodules, they should cite this just-published paper: http://www.amjbot.org/citmgr?gca=amjbot;ajb.1700165v1

2) You will need to explain/contrast sanctions and PFF clearly in Intro to give readers sufficient background.

3) A clarification of "cost".

Reviewer #2

Cost definitions are key. In the model, the cost is necessarily a fitness cost because it is translating fixation rate into population growth rate (i.e., per capita fitness). This is, as articulated in the response to reviewers’ letter, completely distinct from the actual metabolic cost of nitrogen fixation in terms of ATP and reducing power. I don't see the point is having separate terms for fitness cost of fixation (c) and then the fitness effect of sanctions (s1) since these are tied together in this system anyways. It would be clearer to just have a "net fitness effect" of fixation ability that could just be zero, because the growth rate of the fix+ and fix- is the same early in the interaction. The distinction between early phase vs late-phase seems arbitrary.

Related to this, when do Mimosa nodules really start fixing nitrogen? Prior to this point one wouldn't expect there to be a difference between fix+ and fix- since the trait isn't expressed. In Figure 7, it isn't clear whether the first time-point (7d) is significantly different from zero – this would be worth presenting.

Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”: Should specify here that you mean the fitness cost; also, the double negative "preventing mutualism to fail", is confusing and should be revised to "enabling mutualism to spread".

It should be clarified which cells are contributing to the population when nodules are crushed and plated – are these a mix of former bacteroids (the ~20% that are supposedly culturable–so what is happening to the other 80%) and cells from the infection threads? If most of the cells come from infection threads, this could dilute the effect of plant control unless the plant is able to also regulate cells in ITs on the basis of the neighboring bacteroids… Please clarify what you think is happening in this system, as this could guide the next steps in dissecting this phenomenon.

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: Should specify that you are qualitatively testing the model.

Discussion section: degeneration

Discussion section: dashes don't match

Methods: tetracyclin –> tetracycline

Reviewer #3:

I continue to favor publishing the empirical results, while remaining concerned about the distinction between metabolic cost and opportunity cost.

If, as the authors state that in their response, they "have no evidence that nifH mutants can invest more resources in their own fitness," then their results are only relevant to cases where there is no opportunity cost (in terms of potential rhizobial fitness, not just metabolism) to fixing nitrogen. For the Abstract to accurately describe their results, it would have to read something like "rare fixers will invade a population of nonfixing bacteria that lack mechanisms to divert resources from nitrogen fixation to their own fitness…" There's plenty of fluff in the Abstract that could be cut to include this key qualification.

Similarly, where they speculate that "the metabolic cost paid by bacteria to fix nitrogen is too low to be detected" the issue isn't metabolic cost but opportunity cost. Even if metabolic cost is very high, the resources are supplied by the plant. And, if diverting some of those resources to rhizobial reproduction isn't possible, then metabolic costs are irrelevant to rhizobial fitness. That is apparently the case for their nifH mutant. The lack of difference at 14 dpi would only be "surprising" if nifH mutants were able to divert resources to their own reproduction, prior to the imposition of sanctions.

The real question, though, is whether strains that fix less nitrogen and can divert resources saved to their own reproduction can out-compete strains that fix more nitrogen. If we accept the importance of this question, then the empirical results merit publication, because the sanctions shown in Figure 1 are severe enough to outweigh any likely fitness benefits of diverting resources from nitrogen fixation. They don't have any data on the fitness effects of fixing less nitrogen, rather than none – this might not trigger sanctions – but that's not a reason not to publish. It's just a reason not to over-generalize in Abstract and main text. For example, in the Discussion section "in our system" is too vague, especially when coupled with generalizations about "the absence of a social dilemma." That would be a good place to point out that results could be different for "cheaters" (as opposed to "losers"), that is, strains that fix some nitrogen, but divert more resources to their own reproduction than other strains do.

I accept the argument that a model that only applies to small experimental evolution studies (because of low plant numbers) is potentially useful, but the reference to "ecological factors" in the Abstract promises too much.

https://doi.org/10.7554/eLife.28683.034

Author response

Summary:

Baubech et al. examined how legumes deal with rhizobium that does not fix nitrogen. We find parts of the paper very interesting. However, we have problems with your data interpretation, especially when you regard nitrogen fixation as cost-free in your model and in your text. Seems that your results are consistent with partner-fidelity feedback control of non-fixing bacteria (i.e. host cells with fix- mutants don't do well, and thus fix- mutant do not do well). We encourage you to rethink and rewrite.

Reviewer #1:

My feelings for this paper are mixed. I like how the authors showed that nitrogen-starved plants had increased nodules over time (presumably until plant's nitrogen need is met). In a single nodule, fix+ and fix- nodule cells are spatially partitioned, and fix+ increases its frequency over fix-. Fix+ can increase in frequency during several rounds of inoculation/growth, and this increase is faster in longer cycle (as predicted by theoretical work decades ago). However, the paper made quite problematic assertions, which rendered the model and the entire paper problematic.

Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”: "Since non-fixers do not proliferate better than fixers (Figure 1A), they are not cheaters in our system following a fitness-based definition of the term "cheater". This statement (and the thinking behind it, including assuming that nitrogen-fixation has zero cost in modeling) is problematic. You are dealing with a spatially-structured environment. Given the spatial segregation of fix+ and fix- cells to different nodule cells, partner fidelity feedback (see Bull and other's work, which is further extended in https://elifesciences.org/articles/10106) cannot be ruled out. In fact, partner fidelity feedback is likely to operate. In other words, plant cells near fix- bacteria do poorly, and thus fix- bacteria do poorly as well. If you stand behind your claim that N2-fixation is cost free, then you must back up this extraordinary claim with exceptionally careful experiments. That is, if you were to mix the two types together so that they are well-mixed in a plant-like environment (which is probably difficult to do and not done here), fix- cells are more fit or less fit than fix+ cells?

First, we need to clarify our use of the terms ‘cheater’ and ‘cost of nitrogen fixation’.

We employed the term ‘cheater’ in the sense defined by Jones et al., 2015, as we believe that this definition is helpful to think about the evolutionary dynamics of rhizobial populations. This definition states that “cheating must increase the fitness of the actor above average fitness in the population and decrease the fitness of the partner below average fitness in the partner population” (Jones, 2015). This definition therefore considers ‘cheating outcome’ rather than ‘cheating actions’. Although we did not measure plant fitness –our analysis is therefore partial in that respect, we did not observe a better proliferation of non-fixers over fixers at analyzed time points, and thus non-fixers do not “behave” as “cheaters”. Nevertheless, we acknowledge our sentence could lead to confusion and have modified the sentence (see below).

In relation to the above definition, we used the term ‘cost of nitrogen fixation’ to quantify the net cost/benefit ratio of mutualism; once more we employed a fitness-based definition instead of focusing on the actual metabolic cost of nitrogen fixation. Because the Fix+/Fix- fitness ratio is never in favor of the non-fixing strain, we set the cost of nitrogen fixation to 0 in the model (and used another parameter (‘sanctions’) to describe the observed fitness cost of non-fixing strains in older nodules). Several reasons could be invoked to explain the fact that we did not detect any fitness advantage for non-fixing strains: 1) there could be really no metabolic cost; this is however unlikely because 16 molecules of ATP are required to reduce 1 molecule of N2, 2) the metabolic cost could be too small or transient to be detected in our measures or 3) the metabolic cost could be compensated for by stabilizing mechanisms even during the early phases of the interaction. We have thus introduced these possibilities by changing the sentence “Since non-fixers do not proliferate better than fixers (Figure 1A), they are not cheaters in our system” into “Surprisingly, non-fixers did not proliferate better than fixers even at 14 dpi (Figure 1A) possibly because the metabolic cost of nitrogen fixation is too low to be detected in our experimental conditions, or because plant sanctions and the cost of nitrogen fixation equilibrate until sanctions become prominent” (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”).

We have also included a more precise definition of ‘cost of nitrogen fixation’ in the description of the model: “We emphasize that our definition of ‘cost of nitrogen fixation’ therefore does not refer to the actual metabolic cost of this process. It rather refers to the net fitness effect acting on nitrogen-fixing strains, which may integrate the actual metabolic cost and potential plant sanction mechanisms that could be at work in the early phases of the interaction and thus compensate for the metabolic cost” (subsection “Mathematical model and simulations”).

In addition, we would like to note that Cupriavidus taiwanensis, like most rhizobia, fix nitrogen only in plant nodule cells that provide the appropriate environment for this: (i) a large bulk of plant photosynthetates is diverted to bacteroids solving the high ATP demand of the nitrogenase enzyme, (ii) the paradox between the extreme sensitivity of nitrogenase to oxygen and the strict (micro) aerobic status of rhizobia is solved in two ways: nodules exhibit an extremely low free oxygen atmosphere and nodule leghemoglobin – a form of the plant ubiquitous hemoglobin- facilitates O2 diffusion to bacteroids. Furthermore, gene expression in bacteroids is completely modified so that their metabolism is entirely dedicated to nitrogen fixation. Nitrogen fixation only occurs after bacteria have entered the root and massively invaded the nodule cells, i.e. in the fixation zone of the nodule but not in infection threads or in the infection zone (process both spatially and temporally regulated). It is thus impossible to evaluate the metabolic cost of nitrogen fixation in a plant-like environment. In free-living conditions, our isogenic Fix+ and Fix- strains are equally fit (but they do not fix in these conditions).

Second, the concept of partner fidelity feedback was indeed not considered in our current manuscript. We thank the reviewer for the input as this prompted us to think in the direction of the exact ongoing mechanisms.

We do not know the mechanisms that drive the death of non-fixers and host cells in the fixation zone. PFF is an attractive model because it has the potential to stabilize mutualistic interactions without having to invoke additional, unknown mechanisms. The analogy with the yucca-pollinator detailed in Shou, 2015 is particularly interesting for us, since we scrutinized the legume-rhizobia interaction at the cellular level: PC at the nodule level could actually result from PFF at the cellular level.

However, several specificities of the legume-rhizobia interaction deserve further attention. Nodule cell and bacterial degeneration in Fix occupied nodules could result from a defense response from the plant side in response to a bacterial signal (or an absence of signal). A scenario of active plant defense response was recently discovered in Medicago truncatula (Wang et al., 2017; Yang et al., 2017) (plant peptides killing specific bacterial genotypes at the intracellular level) and seems to be consistent with PC through conditional response. Moreover, PFF assumes that plant cells (entities) are metabolically insulated. That is, a plant cell has a limited amount of resources, and it will die if bacteria use up all its nutrients, which we believe is unlikely. Nodules are carbon-sink organs, meaning that the plant devotes significant amounts of carbohydrates to the nodules, and nutrients are exchanged between infected and non-infected plant cells either through transporters and/or by diffusing in the apoplast (White et al., 2007). Non-infected cells play a key role in nutrient exchange between the nodule and the root (Godiard et al., 2011). Therefore, the relevance of the cellular entities for mutualism in the legume nodules is debatable.

Given the arguments explained in the above paragraph and the impossibility to test the spatial equivalence of the focal entities, we feel that we cannot decisively assess which mechanism operates in our system and that our results could be interpreted as either PC through conditional response or PFF. Moreover, another potential mechanism could be that the bacteroid status (complete modification of gene expression in the plant environmental conditions) is lethal to bacteria that do not produce a functional nitrogenase system or do not export nitrogen. Such mechanism could possibly be interpreted as PFF at the cellular level.

Since we do not know the molecular mechanisms responsible for the death of non-fixers, we prefer to be cautious and keep the expression “post-infection sanctions against non fixers” in the text (Discussion section), as commonly employed in the legume-rhizobia literature.

Yet, we have now clarified our definition of the term ‘sanction’, in a way that we believe is compatible with both models, and have mentioned the PFF possibility: “These sanctions could occur as defense responses and/or by decreasing nutrient supply to non-fixing bacteroids. Given that Fix- and Fix+ bacteria are spatially segregated within nodules, the latter case could also result from the local degenerescence of nodule cells, and be interpreted as an example of Partner Fidelity-Feeback mechanism occurring at the level of individual cells 41” (Discussion section).

Figure 1A, Figure 3: I understand fix+ and fix- bacteria, but I am confused about fix+ and fix- nodule. What is the definition?

A Fix+ (or Fix-) nodule is a nodule containing Fix+ (or Fix-) bacteria. We have now made this clear in the Y axis of Figure 1A and Figure 3 and in the legend of Figure 1.

Figure 2B/C: I am not convinced: there are no data points before dpi 14, and so growth rate cannot be credibly estimated.

Indeed, we did not collect nodules before 14dpi, because most nodules are very small before 14dpi and there is a risk that sterilization agents damage them and kill nodule bacteria. We agree that this prevents us from obtaining a precise estimation of growth rate and from detecting any potential cost for nitrogen fixation (see discussion above). Nevertheless, we think that a precise estimation of growth rate is not critical for the purpose of our model, since we don’t use parameter values to extrapolate on potential mechanisms responsible for the observed dynamics. We are essentially interested in the population sizes of the different bacteria, and in their capacity to re-infect host plants (on a fine time-step, allowing to take into consideration stochastic events) over multiple inoculation cycles.

Reviewer #2:

This is a significant contribution to the field of rhizobia-legume symbiosis, demonstrating nicely that non-fixing rhizobia are penalized both at the whole nodule level as well as within nodules – a matter of much debate! – and going farther to considering the time-course of this process. The modeling is a nice addition, though not necessarily required for this to be a strong paper. The multigenerational experiments seal the deal, demonstrating that the phenomenon measured within a generation translates to the expected shifts in allele frequencies over time.

I have only two concerns:

First, as this is something I worry about a lot in my own work, to what extent does the surface sterilization process penetrate and kill rhizobia within Mimosa nodules? This could have dramatic effects on the rhizobial populations observed in single-strain nodules that vary in size, which typically correlates with fixation status.

There is indeed a risk that the sterilization agents penetrate Mimosa nodules especially when the nodules are very small. That is why we did not collect very small nodules and sectioned the root far (ca 2mm) from the basis of each nodule (this last information is now added in Materials and methods section).

Although Fix+-containing nodules and Fixcontaining nodules grow differently (Mimosa indeed form indeterminate nodules), both have a thick cortex as observed by cytology and may not be differently impacted by the sterilization. In co-infected nodules, only Fix- strains are sanctioned and the relative Fix+/Fix- fitness is even higher than when comparing single-occupied nodules.

Second, the authors claim in the text that the model fits their experimental data very well, but do not justify this statistically. In fact, looking at Figure 2 and Figure 10, the model doesn't appear to fit all that well for particular timepoints and there doesn't seem to be all that much power in rejecting the model, if this was in fact the intent. Some consideration of what the purpose of the model is would be warranted–I think it is a nice conceptual addition that shows qualitatively that cycle length and plant population size impact the evolution of fix+ from low frequency, but that the quantitative fit is being overstated.

Thanks for pointing this out. We agree with this comment and certainly need to reword our arguments here. The model was developed as a means to extend the analysis and provides a qualitative estimation regarding the outcome of longer and increased number of cycles, and using plant populations of different sizes. We have incorporated a “nodulation” mechanism within the model to bring it closer to this particular system. However, at its heart the model remains a stochastic dynamical system with inherent variability that has been reduced by parameterizing it on experimental data. We have rephrased our claims as to the use of the model in the manuscript as following:

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: We have added ‘While the model is developed as a proof-of-concept, instead of a simple deterministic model we chose to include stochasticity in the nodulation process in order to reflect the variability observed in the experimental data’.

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: ‘Simulation outcomes qualitatively matched the dynamics’.

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: ‘accuracy’ replaced by ‘predictions’.

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: ‘as predicted by the model’ replaced by ‘similar to what observed with the model.’

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: ‘Data from 21 day-cycle experiment also matched the model prediction (Figure 10B)’ replaced by ‘Under 21 day-cycles, both simulations and experiments lead to a slower progression of Fix+ subpopulations (Figure 10B)’.

Discussion section: “Modelling” was replaced by “Our qualitative model”.

Discussion section: “predicted” was replaced by “A general outcome of the model”.

Discussion section: We added: “The model helps explore further combinations of number of cycles, cycle lengths and plant pool sizes to hypothesize the expected time when Fix+ phenotypes arise.”

Reviewer #3:

This paper presents some important data, although some of the interpretation is questionable. The important results are:

1) Host sanctions against a nonfixing rhizobial strain were much more severe than previously reported for other host species (Figure 1A). In the real world, however, nonfixing nodulating strains like that used in the experiments are rare, relative to less-beneficial strains that fix some nitrogen.

We agree that in nature the capacity to fix nitrogen with a legume partner can vary from no N2 fixation to levels equivalent to nitrogen-fed controls. This can result from deficiency in nitrogen fixation per se or deficiency in a previous stage, such as bacteroid differentiation or persistence, which involve different plant responses (Berrabah et al., 2015). For example, different S. meliloti mutants of bacA, involved in terminal differentiation, are differently impaired in N2 fixation (LeVier and Walker, 2001). Some strains of S. meliloti produce peptidases (HrrPs) degrading plant peptides (NCRs) involved in bacteroid differentiation (Price et al., 2015). The bactericidal activity of other plant peptides (NFS1 NFS2) affects the intracellular survival of specific bacterial genotypes, and thus nitrogen fixation (Wang et al., 2017; Yang et al., 2017). Here we were interested in the impact of nitrogen fixation per se, and thus constructed strain specifically impaired in the nitrogenase system as a first step. Construction of strains producing nitrogenase systems with gradual levels of functionality is challenging. Moreover, we can imagine that, following natural HGT of nod-nif genes in a soil bacterium, a few mutations can shift the recipient genome from Fix- to full Fix+ by integrating nif genes in N2 and O2 regulatory circuitries (and here we show that such mutations will increase in frequency in appropriate conditions). In an experimental evolution approach using a plant pathogen having acquired a rhizobial plasmid as ancestor, one mutation was enough to convert the Nod- ancestor into a Nod+ strain (Marchetti et al., 2010), and two mutations were enough to convert the Nod+ clone into a strain able to nicely intracellularly infect nodules (Capela et al., 2017).

Furthermore, there was no evidence that the resources the nifH strain spared by not fixing nitrogen are automatically invested in rhizobial fitness.

Indeed.

Therefore, we cannot conclude (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”) that "low quality rhizobial partners rarely benefit from low investment in mutualism." (If there were a consensus on this, as claimed, this paper would merit publication only if it challenged that consensus.) To test that hypothesis would require comparing two nitrogen-fixing strains, with one of them providing less benefit via a mechanism that would plausibly enhance its own fitness. Examples include strains that (once inside nodules) block nodulation by other rhizobia (Tatsukami, 2016) or strains that divert more resources from N2 fixation to their own reproduction. Nonfixing mutants would not qualify.

A recent meta-analysis (Friesen, 2012) showed that there is no unequivocal example where rhizobial strains increase their own fitness (again, here we are talking about in planta fitness, i.e. ‘cheating outcome’ – see above answer to reviewer 1) while providing less nitrogen to the host, and thus concluded that “low quality partners rarely benefit from low investment in mutualism”. We acknowledge that this idea is debated. We thus modified the sentence into “The absence of a social dilemma provides support for the emerging idea that low quality rhizobial partners rarely benefit from low investment in mutualism” and moved this sentence to the Discussion section. Moreover, we think that our paper brings an additional level of understanding in this process by providing a detailed spatio-temporal analysis of rhizobial fitness.

We do agree that, in our case, we have no evidence that nifH mutants can invest more resources in their own fitness (see discussion on the cost of nitrogen fixation above) and that the suggestion of the reviewer is an attractive experiment. It would indeed help tease out the threshold in nitrogen fixation below which bacterial fitness declines and identify potential fitness gain of lower-quality partners. Theoretically this could be implemented by assuming a better reproductive success for partial fixers than total fixers. Unfortunately, we do not have such isogenic strains to conduct any confirmatory experiments. As far as we understand, the Gib- mutant strain described by Tatsukami, 2016, although fixing less nitrogen, is seemingly not fitter than the WT strain (again from a ‘cheating outcome’ perspective; see Figure 5B), and therefore would not qualify for such an experiment.

In summary, we acknowledge that interactions occurring in natural ecosystems, between genetically diverse rhizobia and host plants, can encompass a wide range of subtle variations with regards to bacterial nitrogen-fixation output and cost/benefit ratio. However, our aim here was to use a reductionist approach to dissect the response of a legume plant to strains deficient in nitrogen fixation per se and not in previous symbiotic stages. Our hope is that the knowledge gained here can be then be used to interpret results from more complex (possibly ‘realistic’) interactions, in order to contribute to a broader and deeper understanding of the legume-rhizobia mutualism.

2) Sanctions were apparently imposed on a nonfixing strain within mixed nodules (Figure 1B). This is arguably the most-important result, so should be featured in the Abstract, along with the "clear sectoring" (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules") that is presumably key to within-nodule discrimination by the host.

We have now mentioned in the Abstract that sanctions are imposed only on non-fixers in mixed nodules. This result has also now been highlighted in the Discussion section.

3) Figure 3 shows no partner choice (host-imposed discrimination during the infection process). This is consistent with past work comparing isogenic strains, but the authors should recognize the possibility of adaptive partner choice by some host plants among some real-world strains whose signals are correlated with their relative benefits.

We mentioned in the Introduction the existence of partner choice mechanisms in the rhizobium-legume symbiosis. However, to our knowledge there is no study demonstrating that nitrogen fixation per se is controlled via a partner choice mechanism.

Given these empirical results, it's obvious that fix+ rhizobia could invade a fix- population under real-world conditions. I don't really see the point to modeling cases with only 20 plants or where plants only live 14 days after rhizobial infection, as either of those would lead to extinction of the host. If the authors want to explore "ecological factors", as claimed, they should explore some more-realistic factors like temperature or soil nitrogen. What about competition between a strain that fixes half as much nitrogen as another, diverting the resources saved to its own reproduction? That's more realistic than a nonfixing strain. Their relative fitness would depend on how the host's threshold for imposing sanctions and on the extent to which resources diverted from nitrogen fixation actually enhance rhizobial fitness. Until we have those data, modeling is almost pointless. If space in the journal is not an issue, they could keep the modeling work, but the Abstract should be rewritten to be less vague and to highlight the empirical results. One sentence in the Abstract on modeling would be plenty.

The model is useful to build hypotheses regarding the effect of parameter values that are currently beyond the reach of the experiment. Furthermore, it is a stochastic model, bringing the inherent variability of the nodulation process into consideration. Although simple, we see that the model performs well qualitatively, suggesting that it encapsulates the major factors affecting bacterial population dynamics during serial inoculation cycles.

Modeling the expansion/extinction of fixers/non-fixers in “low” conditions such as a plant population size of 20 plants is interesting for experimental evolution approaches, as mentioned in the Discussion section. In our lab we are currently evolving 18 parallel lines of bacterial populations on 20 plants each (which means 320 plants in each cycles). It is thus interesting to know when we have to collect the nodules (length of the nodulation cycle) to get better chance to select Fix+ mutants if they happen to arise.

In this paper we are interested in the impact of nitrogen fixation per se. We agree that it would be interesting to compare strains with different N2-fixing abilities, which would result from different capacities of the nitrogenase system and not from other traits (such as infection or persistence that of course affect the ultimate nitrogen fixation trait; see also discussion above). This imposes to construct such strains, which is challenging.

We have modified the abstract according to reviewer’s comments.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for your revision, but we are returning the revision without further review, because we don't think that you have adequately addressed the concerns about the "costs" that play a central role in your modeling effort (or we were not sufficiently clear in our previous decision letter). In your model, the "cost of nitrogen fixation" being zero strains credulity unless you are very specific about what that these costs mean.

In the revised manuscript we used the term "cost" on two occasions:

– In the Results section where we spoke of the "metabolic cost of nitrogen fixation" referring to the energetic cost paid by bacteria to fix nitrogen, independently from any kind of sanction exerted by the plant. We have now specified this in the text by writing “the metabolic cost paid by bacteria to fix nitrogen.”.

– In the Materials and methods section, where we defined the parameter "c" as being the cost of nitrogen fixation in our model. We wrote " In our simulations, the cost of nitrogen fixation c was set to 0 since we did not detect any difference in the populations of Fix- or Fix+ nodule bacteria at 14 dpi" and “We emphasize that our definition of ‘cost of nitrogen fixation’ therefore does not refer to the actual metabolic cost of this process".

We acknowledge that this second and different definition of the "cost of nitrogen fixation" (c) in the Materials and methods is not the same as the one in the main text and agree that it is not satisfactory. We have now requalified "c" as being the actual "cost of metabolic nitrogen fixation” (same definition than in the results) and splitted the plant sanctions in two terms: s1 that describes basal plant sanctions that compensate for the cost of nitrogen fixation (the only form of sanction during the initial phase of the interaction), and s2uds(t) as an additional form of sanctions that occurs in the later phase of the interaction (Materials and methods section). We specified that “In our simulations, we set c = s1 since we experimentally did not detect any difference in the populations of Fix- or Fix+ nodule bacteria at 14 dpi”. Beyond this time point, additional plant sanctions (possibly including partner fidelity-feedback) against the Fix- bacteria were given by s2uds(t)” (Materials and methods section).

Hence, we think that you need to be very specific about how these costs are described; otherwise the modelling effort could be criticized as being a tautology. If we understand your arguments correctly, the "post-PFF/post-sanctioning costs" need to be very small or even negative (benefit of cooperation) for cooperation to persist. Hence the text should be revised to say something like "the cost of cooperation is difficult to determine because such determination requires a well-mixed environment to be free of PFF. Regardless, this cost must have been overcome by PFF and/or sanctioning mechanisms. The cost we refer to is post-PFF/post-sanctioning cost of cooperation, which should be zero or negative, otherwise the mutualisms would have failed".

In the Results section we have added the following sentences: “…non-fixers did not proliferate better than fixers even at 14 dpi (Figure 1A) possibly because the metabolic cost paid by bacteria to fix nitrogen is too low to be detected in our experimental conditions, or because plant sanctions/PFF and the cost of nitrogen fixation equilibrate until sanctions become prominent. The net cost of cooperation, which is the weighted cost of nitrogen fixation by any form of plant control, thus appeared to be zero or negative, preventing mutualism to fail.”.

In the Materials and methods section we have added: “The metabolic cost of nitrogen fixation is difficult to determine because this process only occurs in plant cells and thus cannot be uncoupled from potential plant control mechanisms”.

In addition, we have added in the Discussion section: “The most likely explanation is that the plant exerts a post-infection control of N2-fixation that overcomes the metabolic cost of nitrogen fixation paid by mutualistic bacteria”. The idea “The cost we refer to is post-PFF/post-sanctioning cost of cooperation, which should be zero or negative, otherwise the mutualisms would have failed” is once more given by the sentence: “Since control mechanisms prevent social dilemma – i.e. the possibility that one partner increases its own fitness by decreasing its investment in mutualism – and help cooperation persist, non-fixers do not threaten mutualism in our system”.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Summary:

Essential revisions:

1) Since their most interesting result was apparent sanctions within mixed nodules, they should cite this just-published paper: http://www.amjbot.org/citmgr?gca=amjbot;ajb.1700165v1

Done (Discussion section).

2) You will need to explain/contrast sanctions and PFF clearly in Intro to give readers sufficient background.

A description of partner choice, post-infection sanctions and PFF is now given in the Introduction.

3) A clarification of "cost".

A description of partner choice, post-infection sanctions and PFF is now given in the Introduction.

Reviewer #2:

Cost definitions are key. In the model, the cost is necessarily a fitness cost because it is translating fixation rate into population growth rate (i.e., per capita fitness). This is, as articulated in the response to reviewers’ letter, completely distinct from the actual metabolic cost of nitrogen fixation in terms of ATP and reducing power. I don't see the point is having separate terms for fitness cost of fixation (c) and then the fitness effect of sanctions (s1) since these are tied together in this system anyways. It would be clearer to just have a "net fitness effect" of fixation ability that could just be zero, because the growth rate of the fix+ and fix-, is the same early in the interaction. The distinction between early phase vs late-phase seems arbitrary.

We apologize for the remaining confusion regarding the definition of the term “cost”. We have now clearly distinguished and defined the “net fitness cost” and the “metabolic cost” in the text, the Materials and methods section and Supplementary file 1:

“[T]he metabolic cost paid by bacteria to fix nitrogen in terms of ATP and reducing power” (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”).

“The resulting net fitness cost of cooperation, which is the weighted metabolic cost of nitrogen fixation by any form of plant control…” (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”).

“We emphasize that a net fitness cost of 0 does not necessarily imply that nitrogen fixation does not impose a metabolic burden on the bacteria” (referred to as ‘metabolic cost’ in the Results section). “Instead, this burden, if significant during the early steps of the interaction, may be compensated for by plant control mechanisms acting at a basal level” (Materials and methods section).

To further address the comment of reviewer #2, we suppressed the parameter ‘s1’ from our model and we went back to our original notation, using ‘c’ as a parameter for the net fitness cost. Therefore, we no longer have a parameter specifically referring to the ‘metabolic cost’ in our model. However, we think it is useful to maintain the two definitions in the main text, in order to justify the value of net fitness cost c=0 (subsection “Mathematical model and simulations”). This is intended to address the comment from reviewer #1 during the first revision, who pointed out that setting a cost to 0 for nitrogen fixation was confusing.

Related to this, when do Mimosa nodules really start fixing nitrogen? Prior to this point one wouldn't expect there to be a difference between fix+ and fix- since the trait isn't expressed. In Figure 7, it isn't clear whether the first time-point (7d) is significantly different from zero – this would be worth presenting.

We have now clarified this issue by including two negative controls in Figure 7, i.e. the measurement of ethylene present in the acetylene reactive and in plants inoculated with a Fix- strain. Statistical analysis showed that the ethylene reduced by Fix- bacteria is significantly different from the negative control background from 10 dpi.

Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”: Should specify here that you mean the fitness cost;

Done (now Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”).

Also, the double negative "preventing mutualism to fail", is confusing and should be revised to "enabling mutualism to spread".

Done (Subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules”).

It should be clarified which cells are contributing to the population when nodules are crushed and plated – are these a mix of former bacteroids (the ~20% that are supposedly culturable–so what is happening to the other 80%) and cells from the infection threads? If most of the cells come from infection threads, this could dilute the effect of plant control unless the plant is able to also regulate cells in ITs on the basis of the neighboring bacteroids… Please clarify what you think is happening in this system, as this could guide the next steps in dissecting this phenomenon.

Nodule bacterial population include bacteria in infection threads and bacteroids within nodule cells of the infection and fixation zones. It has been shown that only 20% of the bacteroids of the nitrogen fixation zone can resume growth on plate (Marchetti et al., 2011). The other 80% are not able to resume growth on plates for unknown reasons.

On nodule sections, Fix- bacteroids of the fixation zone appeared dead at 35dpi (see Figure 5), indicative of a strong control in this plant compartment. Intracellular bacteria of the infection zone often appeared still alive at 35 dpi (Figure 5). Bacteria recovered at 35dpi from nodules only colonized by Fix- bacteria (ca 5 106) may thus represent bacteria present in the infection threads and infection zone. This may explain why the uncooperative population does not become extinct within nodules.

This is now clarified in the text (subsection “Evidence for a spatial and temporal control of mutualism in Mimosa nodules” and the Discusssion section) and in Figure 5.

Subsection “Eco-evolutionary dynamics of N2-fixers and non-fixers through serial nodulation cycles”: Should specify that you are qualitatively testing the model.

Done (subsection “Evidence for a spatial and temporal control of mutualism in Mimosanodules”):

Discussion section: degeneration

Done.

Discussion section: dashes don't match

Done.

Methods: tetracyclin –> tetracycline

Done.

Reviewer #3:

I continue to favor publishing the empirical results, while remaining concerned about the distinction between metabolic cost and opportunity cost.

If, as the authors state that in their response, they "have no evidence that nifH mutants can invest more resources in their own fitness," then their results are only relevant to cases where there is no opportunity cost (in terms of potential rhizobial fitness, not just metabolism) to fixing nitrogen. For the Abstract to accurately describe their results, it would have to read something like "rare fixers will invade a population of nonfixing bacteria that lack mechanisms to divert resources from nitrogen fixation to their own fitness…" There's plenty of fluff in the Abstract that could be cut to include this key qualification.

We do not have any experimental evidence that nifH mutants divert resources from nitrogen fixation to their own fitness, even if it is likely. Neither have we evidence that nifH mutants do not divert these resources to their own fitness.

We thus prefer to stay factual and only use the term ‘non-fixer’ that accurately describes our strain.

Similarly, where they speculate that "the metabolic cost paid by bacteria to fix nitrogen is too low to be detected" the issue isn't metabolic cost but opportunity cost. Even if metabolic cost is very high, the resources are supplied by the plant. And, if diverting some of those resources to rhizobial reproduction isn't possible, then metabolic costs are irrelevant to rhizobial fitness. That is apparently the case for their nifH mutant. The lack of difference at 14 dpi would only be "surprising" if nifH mutants were able to divert resources to their own reproduction, prior to the imposition of sanctions.

We removed ‘Surprisingly’ (subsection “Evidence for a spatial and temporal control of mutualism in Mimosanodules”) from the text.

The real question, though, is whether strains that fix less nitrogen and can divert resources saved to their own reproduction can out-compete strains that fix more nitrogen. If we accept the importance of this question, then the empirical results merit publication, because the sanctions shown in Figure 1 are severe enough to outweigh any likely fitness benefits of diverting resources from nitrogen fixation. They don't have any data on the fitness effects of fixing less nitrogen, rather than none – this might not trigger sanctions – but that's not a reason not to publish. It's just a reason not to over-generalize in Abstract and main text. For example, in the Discussion section "in our system" is too vague, especially when coupled with generalizations about "the absence of a social dilemma." That would be a good place to point out that results could be different for "cheaters" (as opposed to "losers"), that is, strains that fix some nitrogen, but divert more resources to their own reproduction than other strains do.

We acknowledge that in nature all intermediates between non-fixing and fully N-fixing strains occur and that their fate may be different. We have added the following sentence (Discussion section):

“Yet the fate of strains able to fix intermediate levels of nitrogen fixation may be different. Monitoring the fitness of strains varying in their nitrogen fixation capacity would provide a more complete picture of mutualism control”.

Moreover, in the following line, we replaced ‘The absence of social dilemma provides support for…” by “Nevertheless, our results provide an additional example…” (Discussion section).

I accept the argument that a model that only applies to small experimental evolution studies (because of low plant numbers) is potentially useful, but the reference to "ecological factors" in the Abstract promises too much.

Although our experiments are performed in the lab, the factors we tested, i.e. the number of plants, the inoculum size and the duration of the interaction, are in essence ecological, as far as we understand the definition of “ecology”: “Ecology is the scientific analysis and study of interactions among organisms and their environment […] Ecology includes the study of interactions that organisms have with each other, other organisms, and with abiotic components of their environment” (Wikipedia).

This term is widely used for interactions studied in the laboratory, e.g. the experimental evolution of E. coli in minimal medium, virus-host or prey-predator systems:

Le Gac et al. Ecological and evolutionary dynamics of coexisting lineages during a long-term experiment with Escherichia coli. PNAS 2012. 109:9487-9492.

Lenski. Experimental evolution and the dynamics of adaptation and genome evolution in microbial populations. ISME J. 2017. 11:2181-2194.

Dennehy JJ. Am. Nat. 2006 Viral Ecology and the Maintenance of Novel Host Use.

Yoshida et al Nature 2003 Rapid evolution drives ecological dynamics in a predator-prey system

https://doi.org/10.7554/eLife.28683.035

Article and author information

Author details

  1. Benoit Daubech

    The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    Contribution
    Investigation, Methodology
    Contributed equally with
    Philippe Remigi
    Competing interests
    No competing interests declared
  2. Philippe Remigi

    New Zealand Institute for Advanced Study, Massey University, Auckland, New Zealand
    Contribution
    Conceptualization, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Benoit Daubech
    Competing interests
    No competing interests declared
    ORCID icon 0000-0001-9023-3788
  3. Ginaini Doin de Moura

    The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  4. Marta Marchetti

    The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  5. Cécile Pouzet

    Fédération de Recherches Agrobiosciences, Interactions et Biodiversité, Plateforme d’Imagerie TRI, CNRS - UPS, Castanet-Tolosan, France
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  6. Marie-Christine Auriac

    1. The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    2. Fédération de Recherches Agrobiosciences, Interactions et Biodiversité, Plateforme d’Imagerie TRI, CNRS - UPS, Castanet-Tolosan, France
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  7. Chaitanya S Gokhale

    Research Group for Theoretical Models of Eco-evolutionary Dynamics, Department of Evolutionary Theory, Max Planck Institute for Evolutionary Biology, Plön, Germany
    Contribution
    Conceptualization, Supervision, Methodology, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon 0000-0002-5749-3665
  8. Catherine Masson-Boivin

    The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    catherine.masson@inra.fr
    Competing interests
    No competing interests declared
    ORCID icon 0000-0002-3506-3808
  9. Delphine Capela

    The Laboratory of Plant-Microbe Interactions, Université de Toulouse, INRA, CNRS, Castanet-Tolosan, France
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Investigation, Methodology, Writing—original draft
    Competing interests
    No competing interests declared

Funding

Agence Nationale de la Recherche (ANR-12-ADAP-0014-01)

  • Marta Marchetti
  • Catherine Masson-Boivin
  • Delphine Capela

Institut National de la Recherche Agronomique

  • Benoit Daubech

Max-Planck-Gesellschaft

  • Chaitanya S Gokhale

Agence Nationale de la Recherche (ANR-16-CE20-0011-01)

  • Marta Marchetti
  • Catherine Masson-Boivin
  • Delphine Capela

Agence Nationale de la Recherche (ANR-10-LABX-41)

  • Benoit Daubech
  • Marta Marchetti
  • Cécile Pouzet
  • Marie-Christine Auriac
  • Catherine Masson-Boivin
  • Delphine Capela

Agence Nationale de la Recherche (ANR-11-IDEX-0002-02)

  • Benoit Daubech
  • Marta Marchetti
  • Cécile Pouzet
  • Marie-Christine Auriac
  • Catherine Masson-Boivin
  • Delphine Capela

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We are grateful to Jacques Batut, Peter Young and Erik Hom for helpful comments on the manuscript. This work was supported by funds from the French National Research Agency (ANR-12-ADAP-0014-01 and ANR-16-CE20-0011-01), and the French Laboratory of Excellence project 'TULIP' (ANR-10-LABX-41; ANR-11-IDEX-0002-02). BD was supported by an INRA-Région Occitanie fellowship. CSG acknowledges support from the Max Planck Society.

Reviewing Editor

  1. Wenying Shou, Reviewing Editor, Fred Hutchinson Cancer Research Center, United States

Publication history

  1. Received: May 17, 2017
  2. Accepted: October 11, 2017
  3. Accepted Manuscript published: October 12, 2017 (version 1)
  4. Version of Record published: November 15, 2017 (version 2)

Copyright

© 2017, Daubech et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 897
    Page views
  • 216
    Downloads
  • 2
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Evolutionary Biology
    Zachary R Lewis, Casey W Dunn
    Insight
    1. Evolutionary Biology
    Fábio K Mendes et al.
    Research Article