|Home | About | Journals | Submit | Contact Us | Français|
Conceived and designed the experiments: JJP GR MEP BCW. Performed the experiments: JJP GR JDVN. Analyzed the data: JJP LCL MEP JDVN BG. Contributed reagents/materials/analysis tools: JJP GR LCL MEP JDVN ZH JZ GLA BCW. Wrote the paper: JJP MEP PC BCW.
Horizontal gene transfer (HGT) plays a major role in speciation and evolution of bacteria and archaea by controlling gene distribution within an environment. However, information that links HGT to a natural community using relevant population-genetics parameters and spatial considerations is scarce. The Great Salt Lake (Utah, USA) provides an excellent model for studying HGT in the context of biogeography because it is a contiguous system with dispersal limitations due to a strong selective salinity gradient. We hypothesize that in spite of the barrier to phylogenetic dispersal, functional characteristics—in the form of HGT—expand beyond phylogenetic limitations due to selective pressure.
To assay the functional genes and microorganisms throughout the GSL, we used a 16S rRNA oligonucleotide microarray (Phylochip) and a functional gene array (GeoChip) to measure biogeographic patterns of nine microbial communities. We found a significant difference in biogeography based on microarray analyses when comparing Sørensen similarity values for presence/absence of function and phylogeny (Student's t-test; p=0.005).
Biogeographic patterns exhibit behavior associated with horizontal gene transfer in that informational genes (16S rRNA) have a lower similarity than functional genes, and functional similarity is positively correlated with lake-wide selective pressure. Specifically, high concentrations of chromium throughout GSL correspond to an average similarity of chromium resistance genes that is 22% higher than taxonomic similarity. This suggests active HGT may be measured at the population level in microbial communities and these biogeographic patterns may serve as a model to study bacteria adaptation and speciation.
Change in community composition with distance, time, and along environmental gradients (β-diversity) provides information about the mechanisms that generate and regulate microbial biodiversity – and provide insight into evolutionary history  and ecosystem function . Although community structure, evolution  and functional diversity  are all influenced by horizontal gene transfer (HGT), HGT is rarely linked to relevant population-genetics parameters and temporospatial considerations . Genome sequence analyses indicate that preferential transfer of genes is strongly correlated with gene function and is a frequent process in microbial evolution  accounting for much of the biodiversity among isolates –. Genome sequence comparisons (nucleotide and dinucleotide frequency; , codon usage bias; –, or Markov analyses; –) demonstrate horizontal gene transfer of individual organisms, however our current view of HGT is incomplete as it lacks blending population genetics, microbial ecology, and biogeography.
Measuring the transfer of functional genes within ecosystems and relating these events to environmental conditions is a substantial challenge –. Spatial distribution models have been applied successfully in microbial ecology –, , , , in some cases shifting the focus of biogeography from the taxonomic level to functional characteristics that enable survival , . This shift provides a foundation for detailed molecular-level analyses within the context of a sound ecological and evolutionary framework that is required for spatially determining the rate and extent of real world physical gene transfer –. To our knowledge, linking the spatial distribution of functional genes with environmental conditions in a contiguous system has never been addressed.
In this study we examined taxonomic and functional biogeography in the context of the selective pressures in the Great Salt Lake, Utah (GSL). GSL is a hypersaline environment where NaCl concentration ranges from near seawater to saturation, with exceptionally high concentrations of sulfate  and heavy metals  throughout the lake. We analyzed the microbial biodiversity and functional potential across nine sites, chosen for extremes in salt concentration, throughout GSL. Because the majority of environmental microbes cannot be cultured with current laboratory techniques, we utilized recent advances in environmental microarray technology to profile the community structure (using the PhyloChip microarray capable of identifying over 8,000 taxa; ) and functional gene characteristics (using the GeoChip microarray capable of identifying over 10,000 genes in 150 different functional groups; ).
In the summer of 2007, 9 water samples were collected from different sites throughout GSL (Figure S1): Rozel Point (RP, salt saturated; 41°25′56.13″N 112°39′48.31″W), Antelope Island (AI, high salt; 41°02′22.37″N 112°16′42.33″W), Farmington Bay (FB moderate salt; 41°03′31.30″N 112°14′04.98″W), USGS site 3510 (35 40°51′11.07″N 112°20′33.11″W cords) in the South Arm at 3 depths (surface, 3510S; 15.3%, 7m, 3510I; 18%, and 9m, 3510DB; 20% salt concentration) and USGS site 2565 (25, 41°06′58.79″N 112°40′48.33″W) in the South Arm at 3 depths (surface 2565S; 15.4%, 7m; 2565I, 23.1%, and 9m, 2565DB; 23.2% salt concentration). Water from the lake sites was collected at various depths either directly (surface samples) or using a peristaltic pump with flexible tubing that was weighted to minimize horizontal drifting. Samples were collected in sterile Nalgene® 4L plastic bottles. Within 6 h of collection, samples were refrigerated at 4°C until processing. This sampling strategy provided points of data for community analysis ranging from near freshwater to salt-saturated brine.
In order to determine prevailing environmental conditions in which microbial communities reside, we measured dissolved oxygen, pH, salinity via electrical conductivity, and temperature. Water chemistry parameters were measured at lake sites during time of sampling using an In-Situ Troll 9500 multiparameter water-quality monitor. The high range specific conductance and standard pH probes were calibrated and verified prior to taking measurements. Additional measurements involving long-term environmental variation are available through USGS records for sites 3510 and 2565.
We optimized protocols for the extraction of community DNA from the hypersaline waters of GSL using a modification of a protocol published by Griffiths et al. . Due to the near-saturated salt concentration, bacteria cannot be isolated from the samples by filtration as salt precipitates clog the filter. As an alternative, one gallon of water collected from GSL was centrifuged (10,000×g, 40 min, 4°C) in a Sorval high speed centrifuge and resuspended in 500µl of modified CTAB (hexadecyltrimethylammonium bromide) extraction buffer (equal volumes of 10% CTAB in 0.7 M NaCl and 240 mM potassium phosphate buffer (pH 8) . Commonly used bead-beating and chloroform procedures were employed to extract DNA . The extracted community DNA was purified by passing it through a Sephacryl® S-300 column. Briefly, the column was constructed by plugging a 5 ml syringe with sterile glass wool, pouring 5 ml of resin suspended in 24% ethanol into the syringe and centrifuging 10 minutes at 1000×g at room temperature. The column was washed twice with sterile ddH20. Samples were added to the column and purified by centrifugation for 10 minutes at 1,000×g at room temperature. We found that use of this column is critical for good resolution of community DNA and for the elimination of PCR inhibitors present in the water collected from GSL. With this protocol, we have successfully extracted archaeal and bacterial DNA from hypersaline environments, including GSL, and used this DNA to amplify 16S rRNA genes by PCR.
To assess microbial diversity and to overcome obstacles of non-cultivability we used a newly developed 16S Phylogenetic Array (Phylochip) containing probes for 8,741 bacterial and archaeal taxa . Hybridization of the PhyloChip is achieved using slightly modified Affymetrix (Santa Clara, CA) protocols (see ref. ). Briefly, the ribosomal 16S gene was amplified by PCR utilizing Bacteria (F: 5′-AGAGTTTGATCCTGGCTCAG-3′, R: 5′-ACGGCT ACCTTGTTAGCACTT-3′) or Archaea (F: 5′-GACGGGCGGTGTGTCA-3′, R: 5′-GCGGATCCGCGGCCGCTGCAGAYC-3′) specific primers. To minimize the primer bias, PCR amplification was performed with a temperature gradient from 48°C to 58°C for the annealing temperature. The PCR products from the different amplification reactions were collected, purified, and quantified. Two hundred ng of 16S amplicon were fragmented by DNaseI digestion for 20 minutes at 25°C. The DNaseI was then inactivated and the fragmented DNA was biotin labeled for 60 minutes at 37°C following the Affymetrix protocol. The labeled DNA was added to Affymetrix hybridization solution and hybridized to a PhyloChip for 16 hours at 48°C rotating at 60 rpm. The chip was washed and stained following the Affymetrix protocol and scanned utilizing an Affymetrix ChiScanner 3000. Intensity values were normalized using Robust Multi-Array normalization .
To determine the functional genomics capabilities of the microbial communities within GSL, we used the GeoChip functional gene array . Extracted community DNA (no amplification step) was labeled with cyanine-5 (Cy-5) dye. Briefly, approximately 2 µg of genomic DNA was denatured for 5 min at 99.9°C in solution with random octamer mix (Invitrogen, Carlsbad, CA, USA) and snap chilled on ice. Following denaturation, 2.5mM dithiothreitol (DTT), 0.25mM dATP, dCTP and dGTP, 0.125mM dTTP, 0.125mM Cy5-dUTP, and 80U Klenow fragment (Invitrogen, Carlsbad, CA, USA) were added. Reaction mixtures were incubated at 37°C for 3 h. Labeled target DNA was purified with a QIAquick PCR kit (Qiagen, Valencia, CA, USA) according to the manufacturer's instructions. Labeled DNA was measured on a ND-1000 spectrophotometer (NanoDrop Technologies, Wilminton, DE) and dried using a speed-vac at 45°C for 45 min. Dried, labeled DNA was resuspended in a solution of 50% formamide, 5×sodium saline citrate, 0.1% sodium dodecyl sulfate, 0.1 µg µl−1 herring sperm DNA and 0.85 mM dithiothreitol and incubated at 95°C for 5 min. Labeled reactions were kept at 60°C until hybridization. Two technical replicates of community DNA hybridizations were performed using a HS4800 Hybridization Station (TECAN US, Durham, NC) and hybridization conditions were followed as indicated elsewhere  with hybridization temperature of 42°C. GeoChip microarrays were scanned using a ProScanArray microarray scanner (PerkinElmer, Boston, MA) as mentioned by Yergeau et al., . Scanned images were analyzed using ImaGene 6.0 software (BioDiscovery, El Segundo, CA, USA) with signals processed as signal to noise ratio >2.0. The phylogenetic and functional microarray data used in this study comply with journal standards and will be made freely available.
Selective pressure was determined by taking the intensity for different groups of functional genes considered relative to the number of gene variants detected in each group . The microarray design contains multiple probes for each gene sequence or each group of homologous sequences. The richness of gene variants (different gene sequences with the same function) detected for each functional group provided evidence of functional redundancy within each spatially distinct community. Similarly, comparison of the Log2 normalized probe intensity for each functional category indicated the relative abundance of each gene. The relative number of gene variants was determined by dividing the number of genes belonging to each functional category by the total number of genes detected . Relative intensity values for each hybridization signal were calculated and ranked according to intensity to allow comparison of relative abundance of genes in each functional category across experimental samples as per Yergeau, et al., (2007). Figure 1A illustrates the model distribution curve of functional genes through different levels of selective pressure using the competitive exclusion model.
Beta diversity estimates were calculated using presence/absence for individual genes grouped into functional categories as well as 16S genes. Because of the nature of the different arrays (phylochip is PCR-based), we restricted biogeographical analyses where direct comparisons were made to presence/absence based on normalized signal intensity for each array type. We used Sørensen's index for dissimilarity (Bray-Curtis or percent dissimilarity):
where, S1=the total number of genes within a specific functional group detected in the first community, S2=the total number of genes within a specific functional group detected in the second community, and c=the number of genes within a specific functional group common to both communities. The Sørensen index ranges from 0 to 1 where 1 indicates completely different communities and 0 indicates identical communities. Comparison of pairwise dissimilarity across GSL was performed using Mantel tests. To assess the significance of the observed number of shared functional genes between communities, and to test the null hypothesis of random assemblage of communities at sites, we resampled from the total functional gene set to construct 10,000 simulated data sets for each sampling site and estimated the number of shared genes in pair wise comparisons. Site-specific resampling was constrained by the total observed number of observed genes at each site. From these simulated data sets a distribution of shared genes for each pair wise comparison was used to generate significance levels for the observed overlap in functional gene sets.
We detected over 5,000 different 16S rRNA gene sequences of diverse microbial taxa from 9 microbial communities analyzed from GSL ranging from approximately 100 community members in the salt-saturated brine of Rozel Point to 2,400 members in the deep brine sediments in the South Arm (sample site 2565). We detected over 4,500 different functional genes in GSL ranging from 227 different functional genes in the salt-saturated RP community to over 3,000 in the interface between the deep brine layer and surface waters (sample site 3510). The total number of functional genes did not correlate with taxonomic richness across all pooled samples (Pearson correlation, n=9, r=0.28), however the fluctuation in dissolved oxygen among South Arm sites (3510 and 2565) is positively correlated with the ratio of functional genes (GeoChip) to taxa (Phylochip) (Pearson correlation, p=0.046, r=0.82) (Table 1).
Using an approach based on the competitive exclusion principle (see methods), we estimated the selective pressure for each functional category by analyzing the distribution of genes within each sample location. Figure 1B indicates the ratio of the relative intensity to relative richness for each functional group in each location examined throughout GSL. Although the ratio for most functions varies lake-wide, carbon fixation and sulfate reduction ratios are low in all locations, and chromium resistance ratios are high in all locations. Examples of curves for 3510 interface site chromium resistance and sulfate reduction genes are demonstrated in Figure S2.
We used Sørensen's β-diversity to delineate regions or transitions of functional genes (GeoChip) throughout GSL and compared these with taxonomic delineations determined using the PhyloChip. Figure 2 shows the pairwise comparison of the similarity matrix of sample locations for 16S rRNA genes as well as individual functional groups such as metal resistance genes (further divided by specific metals), organic biodegradation genes, and chromosomally encoded functions (sulfate reduction, carbon fixation, etc.) relative to all functional genes detected on the functional gene array. Based on randomized simulated data sets the observed similarity of functional genes between sites is significantly greater than expected by chance for all comparisons except those involving the Antelope Island site and the 3510S site (Table S1). Analysis of chromosomally encoded functions (including sulfate reduction genes) show low (not significant) similarity between sites (Mantel, r=0.47, p=0.11) while we found significant biogeographic patterning for metal resistance (Mantel, r=0.53, p=0.04). Comparison of the β-diversity indices for 16S and functional genes indicates that the change in taxonomic diversity and function is significantly different throughout GSL (pairwise Student's t-test, n=36, p=0.005; see Table S2). Sørensen's diversity in relation to geographic distance shows a very weak correlation in both taxonomic and functional genes (Figure S3).
Similarity values for each functional group were normalized to the similarity value for all functional genes and Log2 transformed. This provides information on which functional groups are more similar than others throughout the lake. Figure 3 demonstrates a weak, yet significant correlation between the relative intensity/richness value calculated above and similarity. Spatial variability of selective pressure across different sites (Figure 1B) breaks the premise of competitive exclusion and, as expected, lessens the correlation near the mean of similarity and selective pressure.
In general, higher salt conditions are restrictive to Cyanobacteria, β-proteobacteria, and Bacteroides, and favor Archaea and Thermotoga (Figure S4). We suggest that the variation in functional diversity within these communities may reflect the environmental dynamics associated with each location. Because of its direct link with the functional repertoire, the diversity of function in relation to the diversity of organisms is believed to be closely coupled to the functional complexity and environmental niche of an organism , . Unvarying environmental conditions favor organisms with a narrow functional repertoire of genes (specialists) while variable environmental conditions favor versatile organisms (generalists) with a wide range of functional potential . Ratios of gene richness to phylogenetic richness in two long-term sites (six samples) when compared with USGS abiotic measurements suggest that more versatile organisms (larger relative functional diversity) are found in areas that vary greatly in oxygen concentration (Table 1). Although variations in oxygen are not responsible for driving all genetic diversity, these data suggest that environmental pressures drive functional diversity in GSL and are consistent with metagenome analyses of HGT . Consequently, the distribution and frequency of functional genes throughout different communities provide insight to environmental pressures experienced by these microbial consortia.
The functional gene array provides a powerful tool for studying microbial biogeography  and ecosystem dynamics in various environments . The functional gene array has sufficient resolution at the functional level to demonstrate how changes in environmental conditions affect the functional structure of microbial communities . In addition, it offers some predictive value with respect to estimating enzymatic activities in microbial communities related to gene families, making correlations between gene abundances and ecological significance rather straightforward . Moreover, the number of gene variants detected offers insight to possible functional redundancy among the dominant community members, while absolute hybridization intensity is indicative of relative abundance of genes .
Resource limitation often drives selection through competitive exclusion  where groups more adept at acquisition and more efficient at resource utilization excel, resulting in fewer competitors (as inefficient competitors decline) (Figure 1A). Similarly, the frequency of a specific function reflects its relative importance in an environment  and is used here as an indicator of selective pressure and successful competition. The principle of competitive exclusion is apt here as the conditions of a single limiting resource (substrate) and as assumption of spatially independent communities. As selective pressure increases, the functional redundancy throughout the community declines with an increase in abundance of functionally similar and competitive variants. For example, high concentrations of chromium throughout GSL  provide a selective advantage for organisms containing the most effective chromium resistance strategies. These more efficient mechanisms increase within the population (either as resistant organisms multiply or as genes are duplicated in the population) and ineffective resistance mechanisms disappear due to toxicity of the environment. The ratio of the relative intensity to relative richness (Figure 1B), therefore, provides a metric for the selective pressure throughout GSL. Conversely, the absence of selective pressure allows diversification of genes as less efficient variants pose no threat to fitness. Sulfate concentration in the GSL is extremely high and is not likely a limiting factor in microbial growth . Consequently, there is little selective pressure for more efficient sulfate reduction genes resulting in more variants and no dominant variants. In this case, the relative intensity is low whereas the number of gene variants is high (Figure 1B). Variation in function, presumably via HGT, rather than changing community, is controlling gene distribution within the environment. Beta-diversity describes the change in biodiversity over space, time, or environmental gradients and often provides ecological and evolutionary information on dispersal, speciation processes, and species turnover. Generally, beta-diversity is used to quantify the species change or turnover in order to delineate biotic regions or transitions . In the case of this study, we use beta-diversity (dissimilarity) to quantify the spatial change of functional genes within the environment. Biodiversity studies are often hampered by artifacts associated with sampling ,  which in this case is minimized using array technology. Each array contains probes for about ten thousand genes, and hence a single hybridization can simultaneously survey a good portion of microbial populations . Despite being a closed format that provides information only about the genes present on the microarray, the Phylochip and GeoChip ensure unbiased comparison of microbial communities because each community is tested against the same set of probes . Although the scale makes a difference in conclusions based on biodiversity estimates , both arrays used here are based on the gene-level scale.
In order to treat the two different approaches (one PCR based, one not) cautiously, we looked at the presence/absence for genes and community members. The average similarity decay of 16S rRNA genes is low throughout GSL (Figure 4A), translating into dispersal limitations presumably due to the salinity gradient. The similarity of all functional genes is significantly higher than that of 16S genes, indicating higher dispersal for all functional gene groups analyzed. These observations are comparable with studies that show a difference in the historical rate of gene transfer between informational genes (16S) and operational genes (function) . Within functional groups, the extent of gene transfer is dependent on whether the function is part of the microbial mobilome  or whether it is chromosomally encoded as part of the core genome  with the exception of phage-transferred genes . Consequently, methanogenesis, a function that is known only to exist in Euryarchaeota (i.e. phylogenetically linked) shows similar biogeographic patterns to 16S genes throughout GSL (t-test p=0.46; Table S1); this pattern is significantly different compared to chromium resistance patterns (Figure S5). This suggests that diversity patterns between the two different types of arrays are comparable and that biogeographic patterns of genes are not random nor are they a result of poor representation on the arrays used. The similarity between chromosomally-encoded sulfate reduction  genes across GSL is low, only slightly (6%) higher than the taxonomic similarity throughout GSL (Figure 4B), whereas similarity of plasmid/transposon-based chromium resistance genes  is 22% higher than the taxonomic similarity (Figure 4C). Although more intensive sampling would improve the resolution (see Figure 4), a significant difference in biogeographic patterns is evident.
We compared individual gene variants with their source to determine whether functional gene biogeography is cryptic within taxonomic biogeography or if the presence of highly dominant species would skew the comparison between taxonomic diversity and functional diversity. The chromium resistant gene sequenced from Deinococcus radiodurans R1 was the only chromium resistance gene detected in all samples from GSL; however, no 16S genes corresponding to any member of Deinococcus-Thermus group were detected in 2 of these samples. This suggests that although the chromium resistance genes likely originated from Deinococcus, they correspond to a different group possibly through a transfer event. Additionally, the most dominant chromium resistance genes throughout GSL corresponded to sequences from β-proteobacteria and α-proteobacteria despite inhibition of β-proteobacteria growth by salt . These data suggest dispersal of functional genes that is independent of taxonomic biogeography.
The exchange of genetic material by microorganisms carries important implications for ecology, evolution, biotechnology, and medicine. HGT is an important factor in the evolution of prokaryotes in promoting adaptation to novel environments by allowing the exchange of large amounts of genetic information that increases the fitness of a specific population within an ecological niche – and plays a large role in controlling gene distribution within an environment by controlling the growth of specific populations . The maintenance and dispersal of genetic elements depends on natural selection parameters that change spatially throughout GSL. Although the biogeographic patterns in GSL alone are not enough to imply HGT, the correlation of these patterns with selective pressure and mobility of functional genes (plasmid/transposon vs. chromosomal) throughout these microbial communities suggest that these patterns are not random. Consistent with previous observation , informational genes involved in transcription and translation, such as 16S, exhibit biogeographic patterns indicating very low levels of transfer compared with functional genes. Within functional genes, horizontal gene transfer corresponds to selective pressure. While gene transfer may occur frequently at the cellular level, this study provides the first demonstration of a measurable link between selective pressure and functional biogeography in a natural community and presents a valuable model for tracking and predicting the dispersal of microbial function.
In many cases increased similarity between sites corresponds to higher selective pressure (e.g. chromium) while decreased similarity corresponds to low selective pressure (e.g. sulfate reduction). Although this study is limited due to array-based analyses, similar approaches to metagenome sequencing datasets could provide improved understanding of the frequency and geographic extent of HGT in real-world communities.
In A the observed number of functional genes in each site are shown in the diagonal and the observed overlap is shown in offdiagonal elements. The associated p-values are shown in B. The p-value is based on a distribution of shared genes generated from 10,000 simulated data sets sampling the observed number of functional genes in each community from the total set of 4560 genes and is the probability of the observed overlap given the null hypothesis of random asemblage of site-specific communities.
(0.16 MB PDF)
Site-to-site Sørensen dissimilarity values according to functional groups.
(0.09 MB PDF)
Sample locations along the salinity gradient in Great Salt Lake. Sample sites 3510 and 2565 are USGS collection sites and samples were collected at the surface, deep brine layer, and the interface between surface and deep brine.
(1.01 MB TIF)
Example of curves from functional genes in 3510 surface sample used to determine selective pressure.
(0.18 MB TIF)
Weak correlation between dissimilarity and geographic distance. Circles represent taxonomic genes (solid line is linear regression). Cross hatches represent functional genes (dashed line is linear regression).
(0.14 MB TIF)
Major phylogenetic shifts due to increased salt. Farmington Bay (FB) was used as reference and the Log2 difference in intensity values are averaged (error=standard deviation) to indicate significant shifts due to high salt.
(0.11 MB TIF)
Average similarity of different genes throughout Great Salt Lake. 16S rDNA (phylochip)gene similarity is not significantly different from taxonomic-dependent methane generation (GeoChip). Sulfate reduction (low selective pressure) is not significantly different in lake-wide similarity from taxonomic genes. Chromium (high selective pressure) biogeographic patterns are significantly different, suggesting independence from taxonomy (t-test).
(0.23 MB TIF)
We are grateful to David Naftz and the United States Geological Survey for access to sampling sites, Chris Garrard and Robert Baskin for geospatial images, and Ethan White for statistical input.
Competing Interests: The authors have declared that no competing interests exist.
Funding: Funding for this project was provided by National Science Foundation grant DEB-021212487 to MEP, and a grant from the United States Department of Agriculture CSREES 2006-34526-17001. This project was supported by the Utah Agricultural Experiment Station at Utah State University as journal paper number 8091. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.