Geographic patterns of symbiont abundance and adaptation in native Australian Acacia–rhizobia interactions


*Author to whom correspondence should be addressed: Peter H. Thrall. Tel.: +61 26246 5126. Fax: +61 26246 5249. E-mail:


  • 1The importance of plant–soil interactions in land reclamation, and the management and restoration of functioning native ecosystems, is becoming widely recognized. However, relatively little is known about broad-scale patterns of genetic variation and adaptation in wild plant–soil symbiotic interactions.
  • 2The current study is part of a larger project examining patterns of genetic variation and adaptation in host and symbiont populations across their geographical ranges using two widespread native Australian Acacia spp. (A. salicina, A. stenophylla) and associated populations of rhizobial bacteria.
  • 3A total of 58 sites were characterized with regard to symbiont population sizes, soil chemistry and environmental parameters. Rhizobial abundance was negatively correlated with a small number of soil factors, including nitrogen, and positively correlated with organic carbon and cation exchange capacity.
  • 4There were clear differences between host species in seedling growth responses in glasshouse trials using pots inoculated with native soils. While A. salicina grew equally well with soils from A. stenophylla and A. salicina sites, A. stenophylla grew best when inoculated with its own soils, indicating broad-scale adaptation to its own rhizobia.
  • 5Rhizobial abundance in these soils was also strongly correlated to variation in nodulation and host growth, indicating that abundance may be a good indicator of the relative effectiveness of rhizobial populations.
  • 6The clear differences in specificity and effectiveness of host–symbiont associations, even among related species, suggests that better knowledge of these systems at multiple spatial scales is central to understanding the factors that influence the ecology and evolution of plant and soil communities, and has potential to increase the cost-effectiveness of restoration programs.


The clearing and fragmentation of native perennial vegetation continues to contribute to environmental problems in many areas of the world, including land degradation, loss of agricultural productivity and declining biodiversity. Native perennial cover is crucial, not only for biodiversity conservation, but also for the provision of biophysical services such as beneficial insects, erosion control and soil infiltration. Consequently, restoration of native plant communities and the landscapes they occupy has become a land management imperative. Increasingly detailed understanding of the key roles that dynamical feedbacks between plant and soil communities play in many ecological and evolutionary processes (e.g. Reynolds et al. 2003; Bennett et al. 2006) has highlighted the importance of introducing appropriate beneficial soil symbionts to improve plant establishment and growth, to aid in remediation of degraded soils, or to improve the sustainability of native ecosystems (Requena et al. 2001; Renker et al. 2004; Harris et al. 2005; Rodríguez-Echeverría & Pérez-Fernández 2005; Thrall et al. 2005).

While the diversity and structure of many natural vegetation communities have been broadly characterized, there is still relatively little quantitative research aimed at understanding the key ecological and genetic factors that would help to optimize restoration processes. In particular, there are major gaps in our understanding of how plant and soil communities are genetically structured and the degree to which provenance must be taken into account when using these associations in a restoration context. Current guidelines often invoke the precautionary ‘local’ principle for plants to preclude concerns associated with local adaptation and outbreeding depression. Although intuitively appealing, such guidelines are largely independent of quantitative information about the geographical scale of genetic differentiation, or the extent to which it impacts on ecological and evolutionary processes. Negative effects associated with geographical-scale movement of plant germplasm have been demonstrated in some cases (Fenster & Galloway 2000; Keller et al. 2000; Montalvo & Ellstrand 2001; Quilichini et al. 2001), but too few studies currently exist to enable broad predictions to be made regarding germplasm movement, even for plant species of high restoration value.

Revegetation activities often focus on the introduction of fast-growing woody pioneer species that are important during the early phases of community re-establishment. Many of these species form associations with beneficial soil symbionts (e.g. Alnus spp. and Frankia; Acacia spp. and rhizobial bacteria; mycorrhizal fungi and a wide range of angiosperm hosts). While spatial and genetic data are accumulating for some plant species, there is still relatively little information available regarding spatial distribution and diversity of ecologically important traits (e.g. nitrogen-fixation) within and among soil microbial communities, or how their functionality is influenced by vegetation community structure. Consequently, there is little understanding of the ecological and evolutionary consequences of large-scale movements of these organisms in either agricultural or restoration contexts (Schwartz et al. 2006).

The potential importance of this issue is supported by previous studies of soil symbionts, which have demonstrated substantial variation in genetic and species diversity (e.g. Lafay & Burdon 1998, 2001; Parker & Spoerke 1998; Barrett & Parker 2006), the degree of host specificity, and the level of benefit conferred by different host-symbiont combinations (Burdon et al. 1999; Thrall et al. 2000). However, much of this work has focused on agricultural systems or on interactions occurring in highly disturbed environments. For example, Svenning et al. (1991) found that populations of white clover (Trifolium repens) from northern Norway were better able to nodulate and fix nitrogen with strains of Rhizobium leguminosarum biovar trifolii from the same region than from southern Norway. One of the best long-term studies of a native plant–symbiont association is the work on the legume genus Amphicarpaea and Bradyrhizobium spp. by Parker and colleagues (Parker 1995, 1999; Spoerke et al. 1996; Wilkinson & Parker 1996; Wilkinson et al. 1996; Parker & Spoerke 1998; Parker et al. 2004). These studies have shown extensive variation in symbiotic specificity and genotypic associations at population, regional and geographical spatial scales.

The potential for complex patterns of adaptation and fitness in coevolutionary interactions such as plant–mutualist associations (Thompson 2005) suggests the need for targeted studies which explicitly assess spatial patterns of variation in ecologically important traits that influence host plant fitness. Developing a better understanding of the synergies between soil microbial and vegetation community structure and function has several implications for conservation and restoration. First, it acts as a benchmark against which remnant conservation or restoration success can be measured. Despite a growing understanding of the importance of soil microbial communities, relatively little thought is given to their maintenance as part of management plans to ensure the long-term viability of vegetation communities. Secondly, restoration can be problematic in the new and relatively hostile environments created by habitat destruction, and elucidating spatial patterns of diversity can provide alternative solutions to restore functionality to the landscape, albeit not necessarily by the organisms present prior to vegetation removal. This is especially important given that soil community structure and function can be radically altered when native plants have been cleared for cropping, or where plant community composition has been radically altered through exotic invasions or long-term grazing (e.g. Hawkes et al. 2005; Wardle 2006). As a result, beneficial symbionts may be absent from disturbed soils where revegetation is most crucial.

The current study is part of a larger project evaluating geographical-scale patterns of genetic variation and adaptation in populations of two widespread Australian species of Acacia (A. salicina Lindl., A. stenophylla A. Cunn. ex Benth.) and their associated rhizobial bacteria. The overall goal is to better understand how plant-soil community viability relates to symbiont abundance, genetic and species diversity, and remnant condition. Better knowledge of these interactions, and the implications of provenance issues for sourcing and deployment, will help to increase the cost-effectiveness of large-scale revegetation. This paper examines patterns of rhizobial abundance and N2-fixing effectiveness in relation to climatic (e.g. temperature, rainfall) and soil variables, and host species.

Specific questions being addressed in this study include:

  • 1How does variation in rhizobial abundance relate to host species and population characteristics (e.g. size, density)?
  • 2How does variation in rhizobial abundance and effectiveness relate to environmental variables (e.g. soil chemistry, temperature, precipitation)?
  • 3Is there variation in rhizobial nodulation effectiveness at promoting host growth, and to what extent is this determined by host species (i.e. is there any evidence that host species are adapted to ‘their own soils’)?
  • 4What are the relationships between rhizobial abundance at individual sites and average effectiveness at N2-fixation?


site selection and derivation of environmental variables

Both of the Acacia species included in this study have broad distributions across the Murray Darling Basin in eastern Australia (Fig. 1), representing a range of climates. Acacia stenophylla occurs on the western interior of the basin from the River Murray north. Acacia salicina occurs throughout the basin, extending more into the eastern flanking ranges than A. stenophylla. Location data from geo-referenced herbarium specimen records, made available through Australia's Virtual Herbarium, and state vegetation surveys were obtained and used as a guide for site selection, and to ensure that populations were sampled across the full climatic range of each species. The combined herbaria and survey data sets provided 429 and 291 records of occurrence, respectively, for A. stenophylla and A. salicina.

Figure 1.

Map showing site locations and identification numbers in New South Wales for the two study species: (a) A. salicina sites; (b) A. stenophylla sites. Major rivers and streams within the Murray-Darling Basin (grey shaded area) are shown.

A survey was conducted across a range of agricultural regions throughout New South Wales between August and November 2005. Sites were selected where the number of mature Acacia salicina or A. stenophylla individuals exceeded 20–30 trees and the distance between locations was greater than 30 km. A total of 28 A. salicina sites and 30 A. stenophylla sites were sampled (Fig. 1). In all but four sites, the two host species did not co-occur, and even in these cases, one or the other host species predominated, with the minority species consisting of a few scattered individuals. In addition to soil sampling, information was collected on site condition (e.g. clearing, grazing pressure), vegetation structure and percentage cover, and host population size.

Climate variables at each location were derived from 1 km grid cell resolution continent-wide climate models based on long-term climate data using the ANUCLIM and BIOCLIM packages (Nix 1986; Houlder et al. 2000). ANUCLIM creates estimations of monthly climatic averages for any location in Australia and from these surfaces BIOCLIM calculates a suite of bioclimatic parameters that describe means, extremes and seasonal variation. Climate data at sample locations was extracted using ArcGIS Spatial Analyst software (ESRI, Redlands, CA, USA).

soil collection and analyses

At each site, soil was collected from around plant bases using a 50-mm soil auger, taking care to ensure that the distance between sampling cores was at least 10 m. At each site 15–20 soil cores were taken to a depth of 10 cm, bulked together and left to air dry. Dried soils were thoroughly mixed, passed through a 2-mm sieve to remove stones and large pieces of undecomposed organic matter and stored at room temperature prior to chemical analysis and quantification of rhizobial abundance. Chemical analyses (based on 500-g subsamples from the bulk soil collections) were done by Incitec Pivot Ltd. These included soil pHCa (0.01 m CaCl2) and pHw, electrical conductivity 1 : 5 soil : extractant, Colwell P (0.5 m KCl) 1 : 100 soil : extractant (Rayment & Higginson 1992), organic carbon (Walkley & Black 1934), inorganic nitrogen (Gianello & Bremner 1986) and exchangeable cations (Rayment & Higginson 1992).

estimates of rhizobial abundance

The most probable number (MPN) plant infection test was used to enumerate rhizobia capable of nodulating Acacia seedlings (Slattery & Coventry 1993). Acacia salicina Lindl., Acacia stenophylla A. Cunn. ex Benth. and Macroptilium atropurpureum Mocino & Sesse ex DC (cv Siratro) were used as trap plants. Acacia salicina and A. stenophylla are nodulated by rhizobia specific to acacias that include strains belonging to other rhizobial genera besides Bradyrhizobium (Lafay & Burdon 1998). This was of particular significance in the current study as there is at least some anecdotal evidence that in more arid/stressed soils, native legumes may more frequently associate with Rhizobium spp. than Bradyrhizobium spp. (Barnett & Catt 1991). Macroptilium atropurpureum is considered to be a broadly promiscuous host for a wide variety of rhizobia, especially Bradyrhizobium spp. (Sprent 2001), and it was included to provide an overall assessment of rhizobial abundance.

Test plants were grown in washed vermiculite, moistened with N-free nutrient salt solution (Gibson 1980), in 150 × 20 mm test tubes closed with plugs of flexible polyurethane foam. For each of the 58 sites × 3 test species, there were six five-fold serial soil dilutions with four replicate test plants at each dilution level. The plants were arranged in wooden racks in a glasshouse with a temperature range of 12–25 °C. Test plants were scored for nodulation 5 weeks after inoculation of the dilution series. Dilutions were repeated for any soils that failed to nodulate across the three test hosts.

glasshouse experiments

The glasshouse study was designed to examine differences among host sites in rhizobial nodulation characteristics and effectiveness at promoting host growth, as well as to specifically address the question of whether host species were better able to form symbiotic associations with rhizobia from their own soils vs. those in which the other host species was dominant. Each host species was grown in all 58 soils, plus an N+ treatment (10 mL of H2O containing 0.05% KNO3 was given weekly to each plant in this group) and an N control. Each host species × treatment combination was replicated 15 times in a complete randomized block design (1800 pots) within a single glasshouse.

For the experiment, 8 cm diameter pots were filled to within 10 cm of the top with a 1 : 1 sterilized vermiculite : sand mixture. For each pot 100 g of a given soil was layered over this mixture and then covered with vermiculite : sand to within 1 cm of the top. Acacia salicina and A. stenophylla seed was obtained from the Australian Seed Company (Hazelbrook, NSW, Australia). Acacia salicina seed was pre-treated by nicking with a scalpel, while A. stenophylla seed was pre-treated with boiling water for 1 min and allowed to cool and stand for 24 h. Pre-treated seeds were transferred to germination trays containing sterilized vermiculite : sand and watered daily. Emerging seedlings were planted into pots after 7–10 days, and the soil surface was covered with a layer of polyurethane beads to eliminate splashing and cross-contamination. Plants were grown under standard glasshouse conditions, watered with N-free 1 : 20 diluted McKnight's solution (McKnight 1949) three times weekly, and UV-sterilized tap water otherwise.

Plants were harvested after 13 weeks and above-ground parts were oven-dried and weighed separately. At the time of harvest, plant roots were separated from the soil and the following nodulation characteristics recorded for each pot: (i) presence/absence of nodules, (ii) nodule number (< 10, 10–50, > 50), (iii) nodule functionality based on colour and size (scores ranged from 1, small-non-N2-fixing nodules with white centres, to 5, large nodules with pink/red centres), and (iv) nodule distribution (scores ranged from 1 to 5, with low scores representing plants with nodules distributed all or mostly within the root crown, and high scores representing plants with nodules more broadly distributed throughout the root system). A subset of the control plants formed nodules late in the experiment; these were small and few in number, and only occurred on lateral roots, thus resulting in negligible impact on plant performance.

statistical analyses

Initial univariate analyses of differences in means for the soil chemistry and environmental data sets indicated a number of significant differences between A. salicina and A. stenophylla sites. However, there were also a considerable number of pair-wise correlations among variables within both data sets. To further explore relationships between biophysical variables and the distribution of the two host species (A. salicina, A. stenophylla), we carried out principal component analyses (PCA) on the soil and environmental data sets separately (excluding derived soil variables, i.e. cation exchange capacity, Ca/Mg ratio, Al saturation and sodium percentage of cations). Individual scores for the principal component axes explaining the most variation (eigenvalues > 1) were retained for use in a discriminant analysis. Because there were relatively few correlations between the environmental and soil PCA scores (r < 0.40 in all cases), we combined these data in the discriminant analysis. We initially did a stepwise discriminant analysis, and then used significant variables that remained in the model in a canonical discriminant analysis.

The principal components identified in the soil and environmental PCAs were also used as predictor variables to examine relationships between each of the log-transformed MPN estimates of rhizobial density (i.e. using M. purpureum, A. salicina or A. stenophylla as trap plants), soil chemistry and environmental variation. In this case, the orthogonal scores from the PCA axes were used as predictor variables in stepwise multiple regressions (with backwards removal). In addition, multiple regression was used to investigate the relationships between the individual soil chemistry variables and each of the MPN estimates. While MPN estimates using A. salicina and A. stenophylla were highly correlated (r = 0.85, P < 0.0001), estimates based on M. purpureum were generally uncorrelated with those from either of the native host species. This suggests that despite its reported broad affinity for Bradyrhizobium spp., M. purpureum may be of limited utility for assessing populations of native bacteria and their potential to form effective symbioses with host plants, especially if host specificity is an issue or where a range of symbiont genera may be involved (e.g. Rhizobium). Thus, for analyses of rhizobial density, we largely focus on estimates derived from the native hosts.

Two-way anova was used to examine the main effects of host species (= soil source) and host population size (as classified above) and the interaction between these variables on variation in rhizobial abundance (log-transformed MPN values). For these analyses, estimates of host plant abundance at each site were categorized as follows: less than 100 mature trees, greater than 100 but less than 1000, and greater than 1000. Pearson product-moment correlations were used to characterize the relationship between the log-transformed MPN estimates for each site, and the mean growth and nodulation responses for both A. salicina and A. stenophylla to the soils from those sites.

A mixed model approach was used to investigate variation in host species nodulation and growth responses to rhizobial populations occurring in different soils. An overall analysis using the combined data for both species (but excluding the N+ and N controls) was used to look at variation in species growth responses to soil source (i.e. whether soils were derived from A. salicina or A. stenophylla sites). Individual analyses by host species were carried out to examine variation among sites, and contrast tests were used to test which sites were significantly different from the N+ and N controls.

To characterize differences in the percentage of plants that formed nodules (presence/absence) according to host species, soil source and the interaction between host species and soil, a generalized linear model with a binary response distribution and logit link function was used. For other nodulation-related variables (nodule functionality, nodule number and distribution), a mixed model was used. Stepwise multiple regression was used to examine the relationship between the mean nodulation characteristics for rhizobial populations and mean host growth responses to these soils in the glasshouse trial. All analyses were performed in SAS 8.2 (SAS Institute Inc., Cary, NC, USA).


host species distribution in relation to environment and soil

Initial analysis of variation in soil chemistry and environmental factors using one-way anovas indicated several significant differences between A. salicina and A. stenophylla sites (Tables 1 and 2). The sampled A. salicina sites generally occurred in areas with lower annual mean temperatures and higher precipitation than A. stenophylla sites. Generally, soils where A. stenophylla was present had lower levels of organic carbon and nitrogen, but higher levels of magnesium, sodium and chloride. In addition, the calculated cation exchange capacity was higher for A. stenophylla sites.

Table 1.  Means (and standard errors) for soil chemistry variables by soil source. F-values are from one-way anovas (d.f. = 1, 56). Significance convention: * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001
VariableA. salicinaA. stenophyllaF value
pH (1 : 5 H2O) 6.82 (0.129)   7.22 (0.156)NS
pH (1 : 5 CaCl2) 6.31 (0.143)  6.63 (0.164)NS
Organic carbon (%) 2.23 (0.178)  1.42 (0.124)14.26***
Nitrate nitrogen (mg kg−1)15.72 (2.166)  9.88 (1.586) 4.81*
Sulphate sulphur (mg kg−1)38.32 (3.451)131.14 (83.836)NS
Phosphorus (mg kg−1)39.56 (5.599) 34.17 (4.989)NS
Potassium (Meq 100 g−1) 1.19 (0.080)  1.17 (0.070)NS
Calcium (Meq 100 g−1)11.66 (1.164) 13.88 (0.774)NS
Magnesium (Meq 100 g−1) 3.74 (0.409)  6.43 (0.480)17.89****
Aluminium (Meq 100 g−1) 0.02 (0.007)  0.01 (0.006)NS
Sodium (Meq 100 g−1) 0.34 (0.051)  1.85 (0.699) 4.36*
Chloride (mg kg−1)60.89 12.033201.17 (67.144) 3.96*
Electrical conductivity (dS m−1) 0.21 (0.013)  0.41 (0.133)NS
Cation exchange capacity16.94 1.483 23.35 (1.330)10.41**
Calcium/magesium ratio 4.28 (0.575)  2.52 (0.250) 8.25**
Aluminium saturation (%) 0.136 (0.067)  0.077 (0.037)NS
Sodium % of cations 2.21 (0.383)  6.26 (1.647) 5.40*
Table 2.  Means (and standard errors) for ANUCLIM-derived environmental variables for the two host species sample sites. F-values are from one-way anovas (d.f. = 1, 56). Significance convention: * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. Units of measure: temperature = °Centigrade; seasonality = coefficient of variation; precipitation = mm; radiation = Mj/m2/day
VariableA. salicinaA. stenophyllaF value
Annual mean temperature 16.48 (0.156) 17.43 (0.238)10.84**
Temperature seasonality192.43 (2.105)196.47 (2.177)NS
Mean temperature wettest quarter 17.38 (1.210) 19.63 (1.215)NS
Mean temperature driest quarter 17.04 (1.000) 15.47 (0.873)NS
Mean temperature warmest quarter 23.53 (0.181) 24.63 (0.286)10.33**
Mean temperature coldest quarter  9.44 (0.132) 10.17 (0.163)12.12***
Annual precipitation458.68 (22.126)358.57 (13.171)15.60***
Precipitation seasonality 16.14 (1.274) 17.60 (1.444)NS
Precipitation wettest quarter138.29 (9.487)108.93 (5.634) 7.30**
Precipitation driest quarter 95.11 (3.595) 73.10 (2.560)25.39****
Precipitation warmest quarter127.89 (10.698)101.20 (6.281) 4.78*
Precipitation coldest quarter108.36 (3.051) 83.50 (2.832)35.76****
Annual mean radiation180.25 (0.826)186.23 (1.284)14.89***
Radiation seasonality 36.96 (0.798)  36.40 (0.648) NS

For the PCA based on environmental factors, the first three axes had eigenvalues > 1 and collectively explained 95% of the variation. The first axis (ENV1) was positively correlated with temperature, and negatively related to radiation seasonality. The second axis (ENV2) was positively correlated with precipitation, while the third axis (ENV3) was positively related to seasonal aspects of both temperature and precipitation. The first four axes (explaining 80% of the variation) were retained from the PCA based on soil chemistry factors. The first axis (SOI1) was positively correlated with variables relating to soil salinity (sodium, chloride, electrical conductivity), but also with sulphate sulphur. The second axis (SOI2) was positively correlated with organic carbon, nitrogen and phosphorus, as well as major soil cations such as potassium, calcium and magnesium, thus representing a measure of soil fertility. The third axis (SOI3) was strongly negatively related to soil pH and calcium, but positively related to aluminium, and was thus a measure of soil acidity, which can negatively impact on nodulation (Bordeleau & Prevost 1994). The fourth axis (SOI4) was strongly negatively linked to nitrogen, but positively linked to magnesium and aluminium.

The first two environmental axes from the PCA were shown to be significant in the stepwise discriminant analysis (ENV1, F1,51 = 5.62, P < 0.05; ENV2, F1,51 = 28.68, P < 0.0001), as were the first and fourth soil chemistry axes (SOI1, F1,51 = 8.70, P < 0.01; SOI4, F1,53 = 21.94, P < 0.0001). In addition, the third environmental axis and the third soil axis were marginally significant (ENV3, F1,51 = 3.37, P = 0.07; SOI3, F1,51 = 3.34, P = 0.07) and therefore were retained for the final discriminant analysis. The discriminant functions based on these canonical predictor variables collectively resulted in an average error rate of 10.2%. Correct classification of A. salicina and A. stenophylla sites was 93% and 87%, respectively. Overall, these analyses indicated that A. stenophylla sites were characterized by higher mean temperatures and lower annual precipitation, and that the soils at these sites had higher soil salinity but lower soil nitrogen levels than those where A. salicina was present. The presence of greater soil N in A. salicina sites may be indicative of the fact that sites for this host species were generally located in higher input cropping and grazing areas of temperate eastern NSW, while the majority of A. stenophylla sites were in central western and western NSW where low input pastoral agricultural systems predominate (Fig. 1).

variation in rhizobial abundance in relation to environment and soil

There was considerable between-site variation in estimates of rhizobial density, ranging over several orders of magnitude (Fig. 2). There were several sites where the MPN tests indicated that the number of rhizobia was below the detectable limit (e.g. Gunbar (site 20), Wanganella (site 14), Pretty Pine (site 12); Figs 1a and 2). While a number of sites with low rhizobial numbers also had small host populations (< 100 individuals), this was not always the case (e.g. the A. salicina population at Pretty Pine was estimated as > 1000 individuals). While for many sites the highest rhizobial densities were detected by the promiscuous M. purpureum, there were a number of sites where, in contrast to the two native host species, M. purpureum either failed to nodulate or produced lower MPN estimates (data not shown).

Figure 2.

Most probable number (MPN) of rhizobia g−1 soil (log values) for soils collected from each of the 58 sites, estimated using either Acacia salicina (black bars) or A. stenophylla (grey bars) as the ‘trap’ host species (see Methods for details). (a) Sites where A. salicina was the resident species (see Fig. 1a for site numbers and locations); (b) sites representing A. stenophylla populations (see Fig. 1b for site numbers and locations).

For estimates of rhizobial density obtained using A. salicina, ENV3, SOI1, SOI3 and SOI4 were retained in the stepwise multiple regression but the overall model fit was poor (inline image = 0.10, F4,53 = 2.61, P = 0.045), and only SOI3 (soil acidity) was significant (standardized regression coefficient (β) = −0.268, P < 0.05). For MPNs based on A. stenophylla, the variables retained were ENV1 (temperature), SOI1 (soil salinity) and SOI4 (negatively related to soil nitrogen) (overall model: inline image = 0.19, F3,54 = 5.71, P < 0.01). All three variables were significant (ENV1, β = 0.287, P < 0.05; SOI1, β = 0.278, P < 0.05; SOI4, β = 0.345, P < 0.01).

To further examine the impact of soil chemistry on rhizobial abundance, stepwise multiple regressions were carried out using the raw soil variables and MPN values derived using A. stenophylla as the test species (results were qualitatively similar for A. salicina). Analyses of host population size and soil source (see below) indicated that, when A. stenophylla was used as the test species, the mean rhizobial abundance detected varied significantly for A. salicina and A. stenophylla soils. Thus, stepwise regressions were carried out for the two soil sources separately. For A. salicina soils, only two variables, calcium (β = 0.535, P < 0.01) and chloride (β = 0.308, NS), were retained, and the final regression was a relatively poor fit (inline image = 0.25, F2,25 = 5.58, P < 0.01). For A. stenophylla soils, variables retained in the final model included nitrogen, calcium, sodium and electrical conductivity (overall model, inline image = 0.41, F4,25 = 5.95, P < 0.01). All four predictor variables were significant (nitrogen, β = −0.442, P < 0.01; calcium, β = −0.375, P < 0.05; sodium, β = −1.816, P < 0.05; electrical conductivity, β = 1.982, P < 0.05).

variation in rhizobial nodulation and n2-fixation

The overall mixed model analysis of plant growth responses (above-ground dry weights) to the different soils indicated large effects of host species, soil source (i.e. whether soils were derived from A. salicina or A. stenophylla), and a significant interaction between host species and soil source (Table 3). These results indicate that, while A. salicina grew equally well with rhizobia derived from its own soils as with rhizobia from A. stenophylla soils, A stenophylla grew significantly better with rhizobial populations present in its own soils than those from A. salicina soils (Fig. 3). Mixed model analyses were also carried out separately for the two host species to examine variation in effectiveness of nodulation and N2-fixation among individual rhizobial populations, particularly in relation to the N+ and N controls. There was considerable variation in performance among soils from different sites (Fig. 4). Mean dry weights for individual soils varied from 0.75 g to 6.54 g for A. salicina (overall effect of site: F59,825 = 8.64, P < 0.0001), and from 0.44 g to 5.63 g for A. stenophylla (site effect: F59,815 = 18.07, P < 0.0001). The average effectiveness of the rhizobial populations varied from those that were highly effective (performing significantly better than the N+ controls as indicated by contrast tests) to those that were so ineffective that they could be regarded as essentially parasitic (Fig. 4). It is worth noting that generally the rank order of effectiveness (= host growth) for rhizobial populations was similar for the two host species (Spearman rank order correlation: r = 0.63, P < 0.0001; Fig. 5), indicating that poorly performing rhizobial populations were consistently so.

Table 3.  Variation in the effectiveness of individual rhizobial populations at promoting host plant growth in relation to soil source (whether soil was derived from an A. salicina or an A. stenophylla site) and host species used in the glasshouse study (host, soil and the host–soil interaction were treated as fixed effects, while variation among sites and the site–host interaction were treated as random variables)
EffectNumerator d.f.Denominator d.f.F-ValueP
  • *

    Tested over the site within soil source MS.

  • Tested over the host × site within soil source MS.

  • Tested over the residual MS.

Soil* 1  56 1.45  0.23
Host 1  5610.40  0.002
Host × Soil 1  5610.96  0.002
Site × Soil56  56 4.95< 0.0001
Host × Site × Soil561598 3.90< 0.0001
Figure 3.

Overall mean dry weight (g) of above-ground biomass for A. salicina (Sa) and A. stenophylla (St) by soil source (i.e. whether soil was sampled from an A. salicina site (black bars) or an A. stenophylla site (grey bars)). Standard errors are shown.

Figure 4.

Variation in mean host growth responses to individual soils relative to the N+ controls (i.e. a value of 0 along the y-axis means no difference from the N+ control). Grey bars represent A. salicina soils, and open bars represent A. stenophylla soils. The black bar to the far right-hand side of the graphs is the N- control. (a) Responses of A. salicina; (b) responses of A. stenophylla.

Figure 5.

Overall correlation between A. stenophylla and A salicina growth performance (mean dry weights of above-ground biomass) using data for individual soils (r = 0.64, P < 0.0001). Filled circles are A. stenophylla soils, and open circles are A. salicina soils.

Variation in the effectiveness of rhizobial populations appeared to be generally unrelated to nodulation per se, as the majority of plants in the glasshouse trial formed at least some nodules (mean percentage of plants forming nodules across A. salicina soils: 94.5% for A. salicina, 92.1% for A. stenophylla; mean across A. stenophylla soils: 99.1% for A. salicina, 97.2% for A. stenophylla). The single exception was the soil from Pretty Pine (an A. salicina site for which MPN estimates were effectively zero), where A. stenophylla showed very low levels of nodulation in the glasshouse trial (20%). Despite the generally high levels of nodulation, the binary analysis of nodule presence/absence indicated significant effects of both soil source and host species (Table 4). Overall, a higher percentage of plants formed nodules with rhizobial populations in soils from A. stenophylla sites, and more A. salicina seedlings were nodulated than A. stenophylla. However, the latter effect of host species became non-significant when the results for Pretty Pine were excluded.

Table 4.  Variation in the percentage of plants forming nodules in relation to soil source and host species used in the glasshouse study
Effectd.f.χ2 valueP
Soil1 6.32  0.012
Host128.32< 0.0001
Host × Soil1 1.48  0.224

With regard to other nodulation measures (nodule functionality, number of nodules formed and nodule distribution), the mixed model analyses for the effect of soil source and host species indicated no significant difference in nodule functionality for either variable. However, for both nodule number and distribution, there was a significant interaction between soil source and host species (Table 5). Overall, rhizobial populations from A. salicina soils formed fewer nodules than those from A. stenophylla soils, and these were also less broadly distributed across plant root systems. To examine the relationship between these measures of nodulation and plant performance, the mean nodulation values for each rhizobial population were analysed against mean plant dry weights by multiple regression. The overall regression (including data for both host species) was significant (adj. R2 = 0.47, F2,113 = 52.39, P < 0.0001), but of the individual predictor variables only the effect of nodule distribution was significant (t1 = 6.42, P < 0.0001); this was positively related to average plant dry weights.

Table 5.  Variation in nodulation characteristics of individual rhizobial populations in relation to soil source and host species used in the glasshouse study
EffectNumerator d.f.Denominator d.f.F-valueP
Nodule effectiveness
 Soil1  55.9 0.790.379
 Host11654 0.130.719
 Host × Soil11654 0.090.765
Nodule number
 Soil1  55.717.22< 0.0001
 Host11654 8.140.004
 Host × Soil1165410.460.001
Nodule distribution
 Soil1  55.616.94< 0.0001
 Host11653 1.570.21
 Host × Soil1165311.45< 0.0001

interactions between host population size, rhizobial abundance and effectiveness

For MPN values estimated using either M. purpureum or A. salicina as trap species, there were no differences in rhizobial numbers in relation to soil source (i.e. whether the resident host was A. salicina or A. stenophylla), host population size or the interaction between these variables. In contrast, when A. stenophylla was used as the trap host, significantly more bacteria were recorded in A. stenophylla sites than in A. salicina sites (mean estimated densities for A. salicina sites, 2.47 × 105 g−1 soil; mean densities for A. stenophylla sites, 1.17 × 106 g−1 soil; F1,52 = 11.57, P = 0.001); as for M. purpureum and A. salicina, there was no consistent effect of host population size on rhizobial abundance. Although the interaction term was significant, it was relatively small (F2,52 = 4.19, P = 0.021) and appeared to primarily depend on the inclusion of the data for the Pretty Pine site. Overall, these results are congruent with the glasshouse study, which indicates that rhizobial populations from A. stenophylla sites are more effective at promoting the growth of A. stenophylla than those from A. salicina sites.

Rhizobial population densities (MPNs) estimated using either A. salicina or A. stenophylla were strongly positively correlated with all nodulation measures from the glasshouse trial, as well as with mean plant dry weights (P < 0.0001 for nearly all comparisons). Given the lack of correlation between MPNs estimated using the native hosts and MPN values from M. purpureum, it is perhaps not surprising that there were very few significant correlations between the M. purpureum MPNs and either the nodulation variables or host growth responses.


Associations between plants and soil symbionts are good systems in which to test hypotheses regarding the impact of host species and environmental factors on the ecological and evolutionary dynamics of soil communities (Thrall et al. 2007), particularly as these associations are known to vary in important characteristics such as host specificity and effectiveness with regard to the provision of benefits (Burdon et al. 1999; Parker 1999; Thrall et al. 2000), and can span the continuum from mutualistic to highly parasitic (Johnson et al. 1997; Denison & Kiers 2004; Thompson 2005; Paszkowski 2006).

A major aim of this study was to characterize geographical patterns of variation in host–symbiont interactions using populations of rhizobial bacteria associated with two native Australian woody legumes. In particular, we were interested in quantifying the impact of host species, environmental and soil factors on traits relative to ecological function (e.g. rhizobial abundance, nodulation and N2-fixing capability). In addition to broad surveys across host ranges, we conducted glasshouse studies to address the question of whether there was variation in rhizobial effectiveness at promoting host growth, and to what extent this is determined by host species (i.e. is there any evidence that host species are broadly adapted to ‘their own soils’)? For example, glasshouse trials found that rhizobial strains associated with a given Amphicarpaea bracteata lineage were more effective partners for that lineage than for plants of other lineages, regardless of geographical origin (Wilkinson et al. 1996). This is supported by genetic studies showing that distributions of rhizobial lineages across geographical host ranges were influenced more by patterns of host genetic variation than by physical distances between populations (Parker & Spoerke 1998).

The present study demonstrates considerable variation in the average effectiveness of nitrogen-fixing capability at the population level for rhizobia associated with two widely distributed Acacia spp. Individual rhizobial populations varied from those that were significantly more effective at promoting host growth than controls with added nitrogen, to those that were essentially no better than the nitrogen-free controls. As evidenced by the positive correlation between the growth of A. salicina and A. stenophylla across soils from the 58 sites, ineffective rhizobial populations were consistently so. This variation in effectiveness was itself strongly related to differences in estimated rhizobial abundance among host sites, suggesting that rhizobial abundance in native soils may therefore be a good indicator of average effectiveness. It should be noted that this correlation was not due to a lack of nodulation in the glasshouse experiments (apart from the low nodulation for A. stenophylla when grown in the soil from the Pretty Pine site, more than 97% of the plants produced nodules).

A possible complication with interpretation of ‘whole soil’ experiments (such as the present study) is that individual soil components are not identified. Rather, results reflect the net effects of both mutualistic and pathogenic elements of those communities; in the present context, it is therefore difficult to determine if results specifically reflect rhizobial effectiveness. While the correlation between MPN values and soil effectiveness does not of itself exclude the possibility that soil pathogens could be important, the correlation between rhizobial abundance and nodule functionality suggests that larger rhizobial populations are forming more effective partnerships with hosts (although this does not determine the direction of causality).

Soil and environmental factors were strongly related to the distribution of host species across the sampled sites. The results of the discriminant analysis showed that it was generally possible to distinguish A. salicina and A. stenophylla sites based on a small number of independent PCA variables. Soil chemistry also clearly played a significant role in determining the favourability of local conditions for rhizobial populations. Rhizobial abundance in particular was negatively related to levels of soil nitrogen, but in general was unrelated to host population size (e.g. the MPN tests were unable to detect any native bacteria at Pretty Pine, where there is an extensive population of A. salicina). It is worth noting that the results from the glasshouse trial showed that, even for A. salicina (which did produce nodules), the rhizobia from this site were relatively ineffective. The difference in nodulation between the MPN estimates and the glasshouse results for A. salicina are most likely due to the fact that the amount of soil used in the glasshouse trials was greater than that used in the MPN dilutions. Given the mature age of the majority of the A. salicina trees at Pretty Pine, it is possible that rhizobial activity is deeper in the soil profile than was sampled (e.g. nodulation on the woody legume Prosopis glandulosa has been shown to be much greater at depth than near the surface; Virginia et al. 1986).

Based on evidence from previous studies for host specificity in host–rhizobial interactions (Parker 1995; Burdon et al. 1999; Thrall et al. 2000), we expected to find variation in host responses to different soils. Interestingly, the two Acacia spp. showed quite different growth responses depending on whether soils were derived from A. salicina or A. stenophylla sites. While A. salicina grew equally well with either type of soil, A. stenophylla clearly favoured populations of rhizobia derived from its own soils, indicating some degree of adaptation. Additional support for differences in host specificity can be seen in the MPN data, which also show that seedlings of A. stenophylla detected more bacteria in its own soils than in those from A. salicina sites. Furthermore, results from an independent glasshouse inoculation trial (J. F. Slattery & P. H. Thrall, unpublished data) using these species but a completely different set of 40 individual rhizobial strains derived from a broad range of native legumes showed that the two species nodulated at similar rates (97% vs. 91% for A. salicina and A. stenophylla, respectively), but that only 11 (27.5%) of these strains were effective at promoting growth for A. stenophylla as opposed to 22 (55%) for A. salicina.

Thus, A. stenophylla appears to have a lower ability to interact with a broad range of isolates than A. salicina despite their similar ecologies and overlapping distributions (both are common in riverine and floodplain environments in the Murray Darling Basin; Fig. 1). This variation in host specificity does not appear to relate to differences in nodulation per se, but rather to the average functionality, number and distribution of nodules, and overall growth responses to symbiont populations. Interestingly, earlier studies suggested that variation in host nodulation specificity and nodule function may be under genetic control by the host (Parker & Wilkinson 1997; Parker 1999), although it remains unclear as to what conditions are likely to select for different degrees of specificity. Previous work on a range of Acacia spp. suggested that host species with more restricted distributions might be more likely to interact with a narrower range of symbionts (Thrall et al. 2000). However, given the broad distribution of A. salicina and A. stenophylla, the present study supports the idea that specificity may evolve for a range of reasons. Such variation in host specificity has also been observed for soybeans (Devine et al. 1990), and hosts with narrow specificity may actually be relatively common (Parker 1999). Overall, spatial patterns of adaptation and polymorphisms in specificity are likely to be determined by trade-offs between the benefits of interacting with a broad range of symbiont strains vs. increased performance with a subset of strains, and the evolutionary consequences of such variation for the evolution of symbiont effectiveness in different environments (Thrall et al. 2007).

Given the observed negative relationship between bacterial abundance and N2-fixing effectiveness with levels of soil nitrogen for rhizobial populations, it may be of interest to consider that A. salicina sites were generally located in areas under more intensive agriculture than those in which A. stenophylla was sampled. The latter were generally located more immediately adjacent to streams and in more sparsely inhabited regions where low input grazing was the primary activity. This may explain the significantly higher levels of soil nitrogen found in A. salicina sites. The relationship between soil fertility and bacterial abundance and effectiveness could either be due to bacteria having low genetic variation for ability to tolerate high N environments, or trees being less likely to form symbioses in more productive agricultural environments.

Theoretical studies predict that shifts from mutualistic to parasitic relationships within symbiotic associations are likely to occur along environmental gradients of increasing productivity (Hochberg & van Baalen 1998; Neuhauser & Fargione 2004; Thrall et al. 2007). Additionally, there is some empirical evidence that host plants may also evolve to have different levels of dependence on soil symbionts in response to changes in soil nutrient levels, as shown by studies of the grass Andropogon gerardii and arbuscular mycorrhizal fungi (Schultz et al. 2001). Interestingly, this also suggests that land management practices may have significant impacts on coevolutionary outcomes between plants and soil symbionts (e.g. in intensively farmed regions, one expectation might be that there would be selection for less beneficial symbionts, while the reverse would be predicted for less disturbed low fertility soils; Kiers et al. 2002).

Given the importance of parasitic and mutualistic associations as drivers of ecological function, there is enormous scope for developing broad-scale studies of plant-soil symbiont (mutualists and parasites) interactions as models for addressing a range of community ecology and coevolutionary issues (Thrall et al. 2007). For example, increased understanding of how plant and soil communities interact at multiple spatial scales (and how these interactions are moderated by physical factors) can provide insights into successional dynamics (De Deyn et al. 2003; Reynolds et al. 2003), geographical distribution of host species (Béna et al. 2005), and the invasion of exotic plant species into native communities (Hawkes et al. 2005; Lafay & Burdon 2006; Parker et al. 2006). The large-scale variation in rhizobial effectiveness and host specificity demonstrated in the present study indicates that a better knowledge of these interactions also has the potential to increase the cost-effectiveness of restoration programs, particularly where soil community structure and function has been substantially altered.


We thank the Armytage family (‘Billyanco’) and Tim and Jane Murray (‘Idalia Station’) for allowing us to sample sites on their properties and Bob Godfree for helpful discussion of statistical analyses. Technical support was provided by Luke Bulkeley, Jacqui McKinnon, Lan Li, Shamsul Hoque and Grant Stewart. The financial support of the NSW Environmental Trust is gratefully acknowledged. This paper was improved by the comments of two anonymous referees.