• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Feb 20, 2007; 104(8): 2761–2766.
Published online Feb 12, 2007. doi:  10.1073/pnas.0610671104
PMCID: PMC1815255

Multiscale responses of microbial life to spatial distance and environmental heterogeneity in a patchy ecosystem


Spatial distance (SD) and environmental heterogeneity (EH) are currently thought to represent major factors shaping genetic variation and population abundance, but their relative importance is still poorly understood. Because EH varies at multiple spatial scales, so too are microbial variables expected to vary. The determination of SD × EH interactions at multiple scales is, however, not a trivial exercise, especially when one examines their effects on microbial abundance and genomic similarities. Here we assessed those interactions at all scales perceptible in a patchy environment composed of known plant species and of heterogeneous soil physical and chemical parameters. For free-living, soil-borne Burkholderia ambifaria, genomic similarities responded to most of the spatial scales that the experimental sampling scheme could reveal, despite limited dispersal of the individuals. Species abundance and community composition were, however, responding to much smaller scales more consistent with local responses to EH. Our results suggest that whole-genome similarities may reflect the simultaneous effects of both SD and EH in microbial populations, but the pure effects of each factor only contributed to <2% of the total genetic variation. The large amount of unexplained variation that remains after considering most environmental, spatial, and biological interactions is then posited to be the result of noise introduced by unmeasured environmental and spatial variability, sampling effects, and neutral ecological drift.

Keywords: biogeography, Burkholderia, multivariate analysis, rhizosphere, spatial modeling

Quantifying the effects of the factors that best explain the variation of abundance and diversity of communities and populations is a central goal in ecology (1). Although spatial variation in abundance and diversity of soil organisms has long been considered “noise” in microbiological studies (2), there is now no more doubt that various free-living microbial populations display clear spatial patterns (reviewed in, e.g., refs. 3 and 4). For instance, the finding of positive relationships between genetic dissimilarities and geographic distances (5) and of a taxa–area relationship (3, 6, 7) for microbes are concepts very similar to what is found for eukaryotic organisms (8).

The current theory of prokaryotic biogeography and diversification proposes two factors to explain those variations. (i) Contemporary environmental heterogeneity (EH) has long been considered as a very powerful factor acting on microbial populations (9) because it is well established that most of terrestrial environments, such as the soil matrix (2), are intrinsically heterogeneous, and microbial abundance and genetic makeup often respond quickly to changing environmental conditions, even in very simple experimental systems (10). Thus, niche-based explanations for the environmental variation in abundance and diversity of microbes have long been the paradigm among microbial ecologists. (ii) The effects of historical events on microbial populations have recently been proposed as another factor that structures microbial life (1113). Spatial distance (SD) may thus be seen as a proxy variable that represents differential community dynamics, which are themselves related to past historical events and disturbances (e.g., physical barrier, anthropogenic activity, dispersal history, and past heterogeneity) whose legacies have been maintained because of spatial isolation between populations (14).

Although it is now well accepted that both EH and SD may help create and maintain microbial diversity in terrestrial ecosystems, little is known about their relative contribution and interactions on the intraspecific abundance and diversity of microbes (3). The identification of the spatial scales at which microbial patterns may be explained by underlying processes remains a central issue. That biotic and environmental variations may share a common spatial structure implies that a part of spatial processes must generally be accounted for when examining the effects of EH on biotic variation (15). Although the partitioning of the ecological variation has sometimes been undertaken in previous microbial biogeography studies (e.g., refs. 6 and 11), it is still not clear how much of the microbial variation can be explained when both SD and EH are considered, especially at different levels of taxonomic resolution and at different spatial scales.

Here, we examined the relative contributions of SD and EH to genetic relatedness and abundance of the Burkholderia cepacia complex (BCC) and one of the dominant species of that complex, Burkholderia ambifaria, which is a cosmopolitan, soil-borne, free-living bacterium often associated with soil roots (16). The genomic relatedness of individual bacterial colonies was determined in a spatially explicit sampling scheme across a patchy agricultural landscape spanning a total area of 250 × 150 m. Environmental heterogeneity consisted of varying soil physical and chemical properties and of predefined patches of four homogenous plant species. Our rationale was as follows: if microbial patterns are mostly determined by historical factors, the genetic distance between isolates should correlate more to SD than to EH; and reciprocally, if the patterns are determined by adaptation to local environmental conditions, their genetic distance should correlate more to EH than to SD. Those hypotheses that involved multivariate interactions were further tested by using a recently described spatial decomposition technique to determine all of the spatial scales that were significantly structuring the microbial community under study.

Results and Discussion

Patterns of Intraspecific Diversity and Dispersal.

Root-colonizing B. ambifaria colonies were isolated from genetically homogenous patches of four crop monocultures (corn, wheat, soybean, and alfalfa) in a spatially explicit experimental setup (see Materials and Methods). Plant roots offer natural hot spots where large populations of BCC are generally found (16, 17). However, culturing BCC, and also most prokaryotes, represents a major technical issue. For instance, Miller et al. (18) reported the presence of BCC in >82% of 91 urban soils by culture-independent methods, but BCC colonies could be isolated in only 15% of all samples investigated. The critical BCC isolation step, therefore, was optimized by developing an improved strategy to recover and identify hundreds of BCC isolates from environmental sources (17). Whereas culture-independent techniques may indeed reveal the extent of sequence diversity in environmental samples, obtaining pure cultures of isolates is the only way to better characterize their phenotypes and genotypes and further assess their ecology.

On each plant root, B. ambifaria isolates were highly diversified, as determined by enterobacterial repetitive intergenic consensus (ERIC)-PCR, a highly reproducible, whole-genome fingerprinting method that is easily amenable to high throughput applications (19). Genetic similarities ranged from identical, as expected from clonally reproducing populations, to completely dissimilar (i.e., not sharing any genomic fragments in common) [supporting information (SI) Fig. 3]. The high intraspecific diversity observed at the root scale may have been generated by high levels of mutations, genomic rearrangements, and recombinations (either homologous or by lateral gene transfer) among or between genotypes, or by high levels of immigration from neighboring sites. This diversity would then be maintained by the interactions of multiple putative factors: the root–soil interface (i.e., the rhizosphere) is known to consist of spatial and temporal gradients of resources, which offer a diverse array of microbial niches (20). Because of root penetration in the soil matrix and solute flow along root surfaces, distinct genetic entities may also be spatially dispersed and further separated (21), thus further contributing to maintaining a high microbial diversity in the vicinity of the roots. Interestingly, although soybean or alfalfa plants were not particularly associated with the presence of BCC populations, the few plants that were BCC-positive harbored contrasting levels of B. ambifaria intraspecific diversity (SI Fig. 3). This finding suggests that the presence of those plant species may locally enrich (i.e., by the “rhizosphere” effect; see ref. 22) for certain BCC genotypes that could be better adapted to grow on leguminous plant roots, depending on the spatial distribution of the preexisting diversity of soil bacteria. The big unknown remains the ecological significance of those genotypes: are they ecologically neutral with respect to each other, as has been posited for coastal bacterioplankton communities (23), or do they represent the genotypes best adapted to prevailing environmental conditions? Because of the very complex nature of the rhizosphere environment, it may also be that natural selection occurs at distinct locations at a given time on the root surfaces and that microbial populations, as we sample them, would then be a mixture of ecologically neutral and adapted genotypes.

To better understand how the bacterial population was spatially structured, a total of 99 genotypes were defined out of the 266 B. ambifaria colonies isolated from soil adhering to the roots of four crop plant species (SI Table 3). Although all samples originated from the same study site, which would potentially allow dispersal between neighboring samples, most of the genotypes (84%) were found at a unique sample location (i.e., on one plant), 12% were found in two samples (most of them occurring at distances >90 m), 2% were found in three samples (most involving distances >100 m), and 2% were found in four samples (distances ranging from 5 to 168 m). Dispersal, i.e., the displacement of individuals and subsequent colonization of new sites, was therefore limited. Significant autocorrelations were evidenced for short lag distances (i.e., corresponding to within-patch variation), which indicated that samples spatially closer to each other, regardless of their location in the geographic range, tended to support more similar populations (SI Fig. 4). Noticeably, the highest density of isolates was recorded on corn and wheat-associated soil samples (SI Table 4). That latter crops were present over the whole study area suggests that dispersal was limited not by small population sizes (i.e., there were theoretically enough individuals at each location to be further dispersed to neighboring sites) but by potential barriers to dispersal, such as physical barriers, physiological requirements, ecological constraints (e.g., more intense competition), or limited resource availability between favorable patches, which would be promoted by high EH.

Interestingly, despite limited dispersal and high local diversification, there was a highly significant, positive relationship between intraspecific genetic diversity and SD (Spearman ρ = 0.269; P < 0.001 based on 999 Monte Carlo permutations) (Fig. 1A) and between intraspecific genetic distance and EH (ρ = 0.130; P < 0.001) (Fig. 1B), over the whole extent of the site. Noticeably, that EH was also spatially structured (ρ = 0.435; P < 0.001) (Fig. 1C) justified the need to further take into account the respective effects of SD and EH on biotic variation (see next section). Despite the influence of environmental factors in shaping microbial population structure and intraspecific diversity, the spatial structure, as measured by SD, thus affected microbial populations at much smaller scales than previously thought. Indeed, by comparing several microbial biogeographical studies, Hughes Martiny and coworkers (3) proposed that, at an intermediate spatial scale corresponding to ten to a few thousand kilometers, both SD and EH could jointly exert their effects on microbial diversity. Here, although at much smaller scales, were observed similar patterns, which could originate from the structural complexity and heterogeneity of the habitats considered but which are not rare characteristics in terrestrial ecosystems (2).

Fig. 1.
Relationships between whole-genome similarity, EH, and SD at the intraspecific level. (A and B) Pairwise whole-genome genetic distances as determined by ERIC-PCR analyses were plotted against SD for 266 B. ambifaria isolates (A), and against environmental ...

Cho and Tiedje (5) revealed the existence of biogeographic patterns for free-living Pseudomonas isolated from pristine soils of worldwide origin. In this study, such patterns could also be detected for a closely related bacterial genus, Burkholderia, but at the level of a single site. A taxa–area relationship was also reported for dominant β-proteobacterial populations inhabiting salt-marsh sediments over a site of a few hundred meters based on rRNA gene similarities, and interestingly, diversity patterns were found to be controlled by EH alone (6). It is now well established that different plant species affect the belowground patterning of diversity of soil microbial populations (24, 25), and diversity patterns in soil organisms may in return have strong influence on both plant community structure and growth of individual plants (2). Here, despite human activities and the presence of patches of homogeneous plant species, historical and environmental signatures could still be detected in the genomic similarities of soil-inhabiting bacteria.

Multiscale Analysis of Biological Variation.

Because spatial structures in biological variables may result from several processes acting at different spatial scales (26), a multiscale spatial analysis, called principal coordinates of neighbor matrices (PCNM) (27), was performed. This spatial decomposition method was applied to the geographic coordinates of the samples, which yielded 15 spatial variables that represented all spatial scales that the sampling scheme could perceive (SI Table 5 and SI Fig. 5). The order of the PCNM variables corresponds to a progression from large to smaller spatial scales (28). The presence of a trend in the response data was also examined, because it would indicate a spatial structure at the broadest scale of the study, i.e., that of the whole study area (26, 28).

The Burkholderia community was examined from the species to the individual genotype level by taking into account abundance and diversity variations, as explained by the spatial scales considered (Table 1). Among the most common species representing the BCC, three species were isolated from the sampling site (SI Table 4). Not surprisingly, B. cepacia and Burkholderia cenocepacia were less frequently isolated than the most cosmopolitan B. ambifaria, as already observed in environmental samples worldwide (ref. 16 and data not shown). Significant spatial structures were identified at all biological levels investigated and explained 14–35% of the biological variance (Table 1). That a significant trend was found only for genomic similarities among B. ambifaria isolates indicated that almost half of the biological variation could thus result from broad-scale processes. For the total BCC and species abundances, one or two spatial variables could generally explain ≈10–20% of the biological variation each time. Scale 12 significantly explained biological variation at all hierarchical levels examined. However, the amount of biological variation occurring at this scale decreased when the focus was shifted from community or species variation (14–21%) to within-species variation (8–9%). Further, <1% biological variation could be explained by spatial variation at this scale when genomic similarities were considered. At the within-species level, up to seven spatial variables were associated with changes in genomic similarities, with levels of explained variation higher at large spatial scales (i.e., scales 2 and 5) than at smaller spatial scales (i.e., scales above 7). At those smaller scales, the amount of explained variation was <1% each time, which further supported the idea that variation in genomic similarities may be generated by broad-scale processes.

Table 1.
Percentage of biological variation explained at different spatial scales

Because very little is known about the driving forces affecting microbial diversity in soil, soil variables were chosen by selecting parameters that are associated with soil fertility and soil physical structure and that may thus have the greatest impact on both above- and belowground life. To better understand how the spatial patterns could be explained by environmental heterogeneity, partial standard regression coefficients were calculated for each environmental variable (Table 2). The specific contribution of each environmental variable was thus determined, while taking into account the effects of all other significant variables in the models. The broadest spatial effect, as represented by the trend, could be generated by fluctuations of soil chemical variables (i.e., pH, NO3), soil physical structure (i.e., clay, sand), and the presence of the different plant species. All the environmental variables in the trend model explained >80% of the spatial variation. Whereas environmental parameters that generated smaller spatial patterns (i.e., scales 8–15) remained enigmatic, the first six large spatial scales could be significantly related to environmental parameters with the plant factor generally present. This result was not unexpected, because plant plots were spatially and randomly distributed over the whole study area, and this design was therefore identified by the spatial variables. Modification of soil structure (i.e., clay and sand) could have generated variation at scale 7, whereas Mg and P variations were related to scales 3 and 6, respectively. Summarizing, fluctuations in genomic similarities that mostly responded to broad spatial scales (Table 1) may be caused by broad fluctuations of the soil environment and by the presence of different plants (Table 2), i.e., by a mixture of both past and contemporary environmental conditions, respectively. In contrast, biological variations at the BCC community and species levels were mostly structured at different, shorter spatial scales (Table 1), because of yet undetermined factors (Table 2).

Table 2.
Partial regression coefficients of variables that significantly explain the spatial scales

Variation Partitioning and Taxonomic Resolution.

A variance-partitioning analysis was carried out to disentangle the effects of space, environmental soil parameters, and plant species on BCC abundance and diversity at the community, species, and within-species levels (Fig. 2). A synthetic view of the amount of biotic variation that may be explained by the variation of each factor alone (i.e., when the effects of all other factors are removed), by the interactions of those factors, and by what remains unexplained, was thus determined (Fig. 2A). At the community level, the abundance of the three BCC species detected was mostly influenced by the plant species (27% of the biological variation; P ≤ 0.001, based on 999 Monte Carlo permutations) and to a lesser extent (4% of the total variation) by Mg fluctuations (Fig. 2B). Although a pure spatial effect was not evidenced, interactions between scale 12 and plant presence could explain up to 10% of the biological variation. The variation in abundance of B. ambifaria, the most prevalent BCC species, reflected that of the three BCC species taken together: 33% of the biological variation could be explained by the presence of different plant species (Fig. 2C). Here, however, the interactions of the three factors accounted for >26% of the total variation.

Fig. 2.
Variation partitioning of microbial abundance and diversity into soil environment (E), plant presence (P) and spatial (S) components at different taxonomic resolutions. (A) General outline. Each diagram represents a given biological variation partitioned ...

At the intraspecific level, the presence of individual B. ambifaria genotypes could be related noticeably to significant variation in pure spatial effects (17%) and plant species composition (5%) but not to effects of soil variables or interactions between the three factors (Fig. 2D). A large part (i.e., 78%) of the variation remained unexplained. Variation in abundance of B. ambifaria genotypes could be explained by significant effects of the three factors considered, each of them accounting for 5–9% of the total biological variation (Fig. 2E). Noticeably, that only 0.3% of the total variation was related to the interactions of the three factors indicates that the effects of each factor were mostly independent of each other. Variations in whole-genome similarities (Fig. 2F) were explained principally by the interactions of soil environment and space (i.e., 8%), but no interaction alone accounted for >2% of the total variation, despite being statistically significant (P < 0.05 for each factor taken alone). A significant pure spatial structure in the biological data, i.e., not shared by the measured environmental variables (fraction S in Fig. 2A), may generally originate from several causes, such as historical events and community spatial processes (e.g., biotic interactions and autocorrelation). However, other causes may also be invoked, such as the presence of additional spatially structured environmental or biotic variables or differential population dynamics that were difficult to detect because of dominant environmental effects (14, 26).

At the level of genomic similarities, although eight environmental variables and eight spatial variables significantly contributed to the pure effects of each factor (see the legend to Fig. 2F), altogether, 73% of the intraspecific variation still remained unexplained. This low proportion of explained variation may help elucidate why discrepancies concerning the relative effects of EH and SD on genetic structure were previously reported (3), because although both of them may be significant, they could contribute to only a small fraction of the total biotic variation. Although sampling effects or unmeasured variability may also contribute to the unexplained variation, it is tempting to speculate that the high fraction of unexplained genomic diversity could also be caused by the evolutionary noise produced by ecologically neutral processes of diversification, i.e., through random ecological drift. The latter could indeed result in the coexistence of many ecologically neutral variants until a selective sweep purges the diversity (23, 29), but they would provide little correlation with the measured spatial or environmental variables. Future studies will need to address those interesting hypotheses.

In conclusion, although microbial populations may locally be highly diversified and spatially isolated from each other, both historical and contemporary environmental conditions have left significant legacies on the intraspecific diversity and abundance of individual genotypes. Although many environmental and spatial variables contributed to the effects of those factors, a large amount of biotic variation still remained unexplained. Those observations were in contrast with community- and species-level variations, which displayed smaller spatial-scale variations and were mostly related to more contemporary environmental conditions. Future investigations will need to focus on the functional significance of the observed patterns of diversity to assess whether the niche theory or the neutral theory of diversity best applies to soil bacteria. Such investigations would certainly provide new insights into how functionally predictable ecosystems can be when the microbial world is included in the equation of ecosystem modeling.

Materials and Methods

Site and Sampling.

Rhizospheric soil samples were obtained in mid-July 2003, from the Biodiversity plots of the Long-Term Ecological Research site at the W. K. Kellogg Biological Station (Hickory Corners, MI). The Biodiversity plots consist of four blocks composed of 21 randomly distributed 10 × 27 m plots of different crop species, with each treatment present once in each block. Corn (treatment B18), soybean (treatment B19), wheat (treatment B20), and alfalfa (treatment T6) plots were selected based on the fact that each treatment was managed under crop monoculture (one annual crop with no cover crop and no crop rotation), with conventional tillage, and without application of nitrogen, herbicides, and insecticides. Samples consisting of one complete plant root system per location were taken along linear transects within a given plot from 0 to 20 m, whereas for larger distances, sample locations were determined by plot locations (samples were then taken in the center of the plots). The location of each sample was recorded for spatial analyses. Samples were refrigerated and transported back to the laboratory within 2 h after collection and were further archived at −20°C. For each sample location, four subsamples of the topsoil (5–20 cm) were collected and mixed to determine soil physical and chemical analyses. The analyses of pH, NO3, OM, Ca, Mg, K, P, sand, clay, and silt composition were done by the Soil and Plant Nutrient Laboratory, Michigan State University, according to the recommended chemical soil test procedures for the North Central region (www.css.msu.edu/SoilTesting/).

Bacterial Variables.

BCC presence and abundance were determined by using a combination of culturing on BCC semiselective medium, BCC-specific colony hybridization, and duplex PCR on 1-g samples for each root system, as described in ref. 17. This procedure recovers BCC from the rhizospheric soil fraction, i.e., the soil fraction adhering closely to plant roots and hence under their direct influence (22). The standard deviations of abundance values typically ranged from 0.1 to 0.4 log10 (cfu/g of roots) (SI Tables 3 and 4). Absence of the target populations from each negative sample was confirmed by PCR on total DNA of BCC-enriched samples or on total DNA extracted from the samples without cultivation (17). BCC species status was determined by recA classification of recA-RFLP patterns, followed by confirmatory recA and 16S rRNA gene sequencing for representative isolates (17).

Genotypic similarities within Burkholderia species were obtained by ERIC-PCR, using primers ERIC1R and ERIC2 (19). Genomic fragments ranged on average from 30 to 40 bands for each isolate (SI Fig. 6). Only profiles consisting of >20 genomic bands in a 300- to 3,000-bp size range were included in the analysis. Band sizing was done by comparison with a 1-kb DNA ladder (GIBCO, Carlsbad, CA) (three ladders for each gel row) by using GelCompar II software (Applied Maths, Austin, TX), and pairwise similarities between fingerprints were determined by using Pearson's correlation coefficient, i.e., by incorporating both band presence and intensity for each profile. Based on the experimental reproducibility of the profiles, which were repeated at different occasions and for various subsets of isolates, a conservative definition of unique ERIC-PCR genotypes was used, which consisted of a resolution cutoff of 85% minimal similarity value (SI Fig. 6).

Data Analyses.

When the objective was to correlate EH with SD or with genetic similarities (Fig. 1), Euclidean distances between samples were computed based on standardized values of soil physicochemical parameters (SI Table 4). When the objective was to understand more specifically which environmental variables were significantly correlated with SD or with biological variation, EH consisted of the environmental variables that significantly explained the variation in biological data (see below).

The spatial trend was removed from the data by linear regressions on (x, y) coordinates of all samples of the study site before PCNM (27) analysis was performed. The PCNM method decomposes the total spatial variation into a finite set of explanatory spatial variables, each of which corresponds to a specific spatial structure or scale. Hence, this decomposition enables a spatial analysis at all scales that can be perceived by the sampling scheme (28). The eigenvalue decomposition of the distance matrix was done by using a truncation distance of 37 m, as determined by measuring the largest distance between first neighbors and by adding a supplementary site at coordinates (x = 22 m, y = 10 m) to increase resolution (27). The detrended biological variables were regressed on the PCNM variables, and the significance of the regression coefficients was tested by 999 permutations of the residuals under the full regression model (26). To determine the environmental parameters that significantly explained the spatial patterns at different scales, a forward selection procedure was first applied to all quantitative and qualitative environmental variables (Table 2). The significance of the conditional effects was determined each time by 999 Monte Carlo permutations of the residuals of the full regression models. Based on each selected subset, simple or multivariate linear regressions (i.e., by using redundancy analysis, RDA) were then applied to calculate the total explained variance and the partial regression coefficients for each variable in the model by using the package CANOCO (Microcomputer Power, Ithaca, NY) (30). The following normalizing transformations of the variables were done before performing multivariate analyses: P, K, Ca, Mg, and NO3 were ln(x + 1)-transformed; the percentages of clay, silt, sand, and OM were arcsin (x)-transformed; and pH values were left unchanged. The presence of different plant species was recoded as a set of dummy variables, to include those categorical variables into the analyses (26). For soil texture variables (i.e., sand, clay, and silt) or plant species variables (i.e., corn, wheat, soybean, and alfalfa), all variables of the respective category were retained in the final models when at least one variable was found significant after the forward selection procedure. To analyze the variation of genomic similarities as represented by ERIC distance matrices, distance-based RDA (db-RDA) was used. This method enables a multivariate analysis of distance matrices, as explained by the effects of each factor alone or in combination (31). Multiple regressions, RDA, db-RDA, and partial constrained analyses (to control for the effects of specific factors or variables in the analyses) were also performed with CANOCO.

The spatial and environmental variations in biological data were further examined by canonical variation partitioning (15) as follows. Based on the set of significant variables in each variable category (i.e., soil physical and chemical parameters, presence of plant species, and spatial variables), as determined by forward selection on the explanatory variables, variation in microbial data (abundance, diversity, or dissimilarities) was partitioned into the pure effects of those factors and into their interactions (15, 26) by using unbiased estimators of the fractions (32). For each of the standard or partial canonical methods used, the significance of the testable fractions (i.e., explained fractions of variation accounted for by the sum of the canonical axes) was determined by using 999 Monte Carlo permutations of the residuals under the full model (30). Two methods have traditionally been used to partition the variation of community composition data, i.e., canonical partitioning and regression on distance matrices based on Mantel tests (26). The canonical approach has been shown to be more appropriate than the Mantel test approach to correctly partition the beta diversity among sites and to test hypotheses about the origin and maintenance of its variation (33).

Supplementary Material

Supporting Information:


We thank Brendan J. M. Bohannan and John W. Taylor for constructive comments on the manuscript, Pierre Legendre for helpful comments about statistical analyses, Eric F. Zorn and Melissa S. Meyers for excellent technical assistance, and Andrew T. Corbin for help with sampling at the W. K. Kellogg Biological Station. This work was supported by U.S. National Science Foundation Grants DEB 0075564 and DEB 05162252 (to J.M.T.) and a postdoctoral fellowship from the Swiss National Science Foundation (to A.R.).


Burkholderia cepacia complex
environmental heterogeneity
enterobacterial repetitive intergenic consensus
organic matter
principal coordinates of neighbor matrices
spatial distance.


The authors declare no conflict of interest.

This article contains supporting information online at www.pnas.org/cgi/content/full/0610671104/DC1.


1. Tuomisto H, Ruokolainen K, Yli-Halla M. Science. 2003;299:241–244. [PubMed]
2. Ettema CH, Wardle DA. Trends Ecol Evol. 2002;17:177–183.
3. Hughes Martiny JB, Bohannan BJM, Brown JH, Colwell RK, Fuhrman JA, Green JL, Horner-Devine MC, Kane M, Krumins JA, Kuske CR, et al. Nat Rev Microbiol. 2006;4:102–112. [PubMed]
4. Ramette A, Tiedje JM. Microb Ecol. 2006 doi: 10.1007/s00248-005-5010-2. [PubMed] [Cross Ref]
5. Cho JC, Tiedje JM. Appl Environ Microbiol. 2000;66:5448–5456. [PMC free article] [PubMed]
6. Horner-Devine MC, Lage M, Hughes JB, Bohannan BJM. Nature. 2004;432:750–753. [PubMed]
7. van der Gast CJ, Lilley AK, Ager D, Thompson IP. Environ Microbiol. 2005;7:1220–1226. [PubMed]
8. Horner-Devine MC, Carney KM, Bohannan BJM. Proc R Soc London Ser B. 2004;271:113–122. [PMC free article] [PubMed]
9. McArthur JV, Kovacic DA, Smith MH. Proc Natl Acad Sci USA. 1988;85:9621–9624. [PMC free article] [PubMed]
10. Travisano M, Rainey PB. Am Nat. 2000;156:S35–S44.
11. Whitaker RJ, Grogan DW, Taylor JW. Science. 2003;301:976–978. [PubMed]
12. Papke RT, Ramsing NB, Bateson MM, Ward DM. Environ Microbiol. 2003;5:650–659. [PubMed]
13. Papke RT, Ward DM. FEMS Microbiol Ecol. 2004;48:293–303. [PubMed]
14. Borcard D, Legendre P. Environ Ecol Stat. 1994;1:37–61.
15. Borcard D, Legendre P, Drapeau P. Ecology. 1992;73:1045–1055.
16. Parke JL, Gurian-Sherman D. Annu Rev Phytopathol. 2001;39:225–258. [PubMed]
17. Ramette A, LiPuma JJ, Tiedje JM. Appl Environ Microbiol. 2005;71:1193–1201. [PMC free article] [PubMed]
18. Miller SCM, LiPuma JJ, Parke JL. Appl Environ Microbiol. 2002;68:3750–3758. [PMC free article] [PubMed]
19. Rademaker JLW, Louws FJ, de Bruijn FJ. In: Molecular Microbial Ecology Manual. Akkermans ADL, van Elsas JD, de Bruijn FJ, editors. Dordrecht, The Netherlands: Kluwer; 1998. pp. 1–27. Vol 3.4.3.
20. Bais HP, Weir TL, Perry LG, Gilroy S, Vivanco JM. Annu Rev Plant Biol. 2006;57:233–266. [PubMed]
21. Watt M, Silk WK, Passioura JB. Ann Bot. 2006;97:839–855. [PMC free article] [PubMed]
22. Hiltner L. Arb Dtsch Landwirt Ges. 1904;98:59–78.
23. Acinas SG, Klepac-Ceraj V, Hunt DE, Pharino C, Ceraj I, Distel DL, Polz MF. Nature. 2004;430:551–554. [PubMed]
24. Saetre P, Baath E. Soil Biol Biochem. 2000;32:909–917.
25. Garbeva P, Postma J, van Veen JA, van Elsas JD. Environ Microbiol. 2006;8:233–246. [PubMed]
26. Legendre P, Legendre L. Numerical Ecology. Amsterdam: Elsevier Science; 1998.
27. Borcard D, Legendre P. Ecol Modell. 2002;153:51–68.
28. Borcard D, Legendre P, Avois-Jacquet C, Tuomisto H. Ecology. 2004;85:1826–1832.
29. Cohan FM. In: Microbial Genomes. Fraser CM, Reed TD, Nelson KE, editors. Totowa, NJ: Humana; 2004. pp. 175–194.
30. ter Braak CJF, Smilauer P. CANOCO Reference Manual and CanoDraw for Windows User's Guide: Software for Canonical Community Ordination. Ithaca, NY: Microcomputer Power; 2002. (Version 4.5)
31. Legendre P, Anderson MJ. Ecol Monogr. 1999;69:1–24.
32. Peres-Neto PR, Legendre P, Dray S, Borcard D. Ecology. 2006;87:2614–2625. [PubMed]
33. Legendre P, Borcard D, Peres-Neto PR. Ecol Monogr. 2005;75:435–450.

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...