Environmental controls over bacterial communities in polar desert soils

Productivity-diversity theory has proven informative to many investigations seeking to understand drivers of spatial patterns in biotic communities and relationships between resource availability and community structure documented for a wide variety of taxa. For soil bacteria, availability of organic matter is one such resource known to influence diversity and community structure. Here we describe the influence of environmental gradients on soil bacterial communities of the McMurdo Dry Valleys, Antarctica, a model ecosystem that hosts simple, microbially-dominated foodwebs believed to be primarily structured by abiotic drivers such as water, organic matter, pH, and electrical conductivity. We sampled 48 locations exhibiting orders of magnitude ranges in primary production and soil geochemistry (pH and electrical conductivity) over local and regional scales. Our findings show that environmental gradients imposed by cryptogam productivity and regional variation in geochemistry influence the diversity and structure of soil bacterial communities. Responses of soil bacterial richness to carbon content illustrate a productivity-diversity relationship, while bacterial community structure primarily responds to soil pH and electrical conductivity. This diversity response to resource availability and a community structure response to environmental severity suggests a need for careful consideration of how microbial communities and associated functions may respond to shifting environmental conditions resulting from human activity and climate variability.


INTRODUCTION
Despite early suggestions that microbial taxa experience cosmopolitan distribution (Baas Becking 1934, Finlay 2002), recent evidence is reveal-ing a diverse microbial world exhibiting spatial patterns over environmental gradients spanning meter, kilometer (Noguez et al. 2005, Zeglin et al. 2011) and regional to continental scales (Fierer and Jackson 2006, Yergeau et al. 2007, Bryant et al. 2008).Although geographic patterns in microbial communities are now evident, the mechanisms driving them remain poorly understood (Soininen 2012).One promising avenue for beginning to frame hypotheses behind microbial biogeography is the application of macroecological theory, with which researchers may test whether controls over the spatial organization of eukaryotic communities are equally appropriate for microorganisms (Martiny et al. 2006, Soininen 2012).
Environmental controls have been long recognized as major drivers of a species' presence/ absence and abundance (Ricklefs and Schluter 1994).For instance, productivity-diversity hypotheses predict that spatial or temporal variation in resource availability (e.g., nitrogen, organic matter) influences communities by eliciting niche-specialization of, and even competitive exclusion by, particular species across ranges of nutrient availability or productivity (Tilman 1982, Waide et al. 1999).Often a unimodal (hump-shaped curve) relationship is observed between productivity and diversity, a result of positive effects of resource availability and negative effects of competition along a gradient of increasing ecosystem production (Michalet et al. 2006).Field surveys (Abramsky andRosenzweig 1984, Mittelbach et al. 2001) and experimental resource manipulations (Silvertown et al. 2006, Chase 2010) support these predictions for a variety of taxa in terrestrial and aquatic ecosystems.Environmental severity (e.g., extreme pH or salinity) may also generate a range of physical conditions that influences productivity, habitat suitability, and community structure of organisms (Freckman andVirginia 1997, Lee et al. 2012).Given the universal constraints to biological diversity and activity which arise from both resource limitations and environmental severity, these mechanisms likely operate together to control macro-and microorganismal community structure alike (Prosser 2007).Indeed, examples of significant productivity/diversity relationships have been reported for microorganisms (Horner-Devine et al. 2003, Smith 2007, Logue et al. 2012).
Antarctica's polar deserts are a model system in which to address questions of controls over microbial biogeography.Resident organisms are sensitive to resource availability, and abiotic factors in general, given the exceptionally low soil organic matter content and extreme ranges of soil pH (.9.0) and salinity (.10,000 lS/cm) (Bockheim 1997).Distinct biogeochemical gradients span orders of magnitude in nutrient availability, major ion concentrations, and biomass (Barrett et al. 2004, Poage et al. 2008), characteristics of a landscape where abiotic factors are the primary controls over the diversity and structure of microbial communities.Recent research conducted in this region has described responses of microbial communities to a number of abiotic drivers including: water availability (Zeglin et al. 2011); geochemistry (Lee et al. 2012); carbon concentration (Aislabie et al. 2009); or a combination of these factors (Niederberger et al. 2008, Smith et al. 2010, Stomeo et al. 2012).The anticipated response of dry valley biota to environmental controls is also supported by evidence from other ecosystems and experimental manipulations, which have demonstrated a community response to environmental variability, such as positive influences of moisture levels on diversity (Zhou et al. 2002) and effects on community similarity due to carbon substrate (resource) diversity (Orwin et al. 2006, Eilers et al. 2010).
Here we focus on the distinct effects of resource availability on both bacterial community diversity and structure, as well as the influences of geochemical severity (pH and salinity).To evaluate these abiotic drivers we studied soils representing a productivity gradient while additionally capturing an extensive range in geochemical conditions.Based on evidence from productivity/diversity theory and other field surveys, we hypothesize: (1) bacterial community diversity will exhibit a positive relationship with resource availability (organic carbon and/or water) and a negative association with geochemical severity, while (2) bacterial community structure will be influenced by both resource availability and geochemical severity.In addition we quantify a gradient of primary production (chlorophyll a) for Antarctic soils and examine the influence of aboveground productivity on belowground biology and biogeochemistry.

Site description
The McMurdo Dry Valleys are an ice-free polar desert in Southern Victoria Land, Antarctica.Aridity (generally less than 10cm of annual precipitation), temperature (mean annual between À168C to À218C), and low soil organic carbon availability (0.03% average by weight) together constrain the diversity and activity of native biota (Kennedy 1993, Burkins et al. 2000, Hogg et al. 2006).Dry permafrost soils are poorly weathered and composed of .90%sand-sized particles with ice cement occurring within 0.5 m of the surface (Ugolini and Bockheim 2008).Salinity and pH are generally high, a consequence of limited vertical water movement through soil layers that results in the accumulation of weathered carbonates and aerially-deposited salts, particularly on old, exposed surfaces (Bockheim 1997).Only during the austral summer (November-February) does 24-hour incident radiation generate above-freezing temperatures, inducing melt and stimulating biological activity (Fountain et al. 1999, McKnight et al. 2007).Liquid water creates environmental gradients across fine and landscape scales by altering soil geochemistry, e.g., environmental severity gradi-ents (Barrett et al. 2009) and promoting localized hotspots of primary production (Barrett et al. 2006).
The soil food web is simple, microbiallydominated, and at the base composed of various prokaryotic, photosynthetic bacteria (Families Nostocaceae, Oscillatoriaceae), eukaryotic algae (Phyla Chlorophyta, Bacillariophyta), and fewer than 10 species of moss (Family Bryaceae) (Broady 1996, Seppelt andGreen 1998) that associate in cryptogamic communities.Several species each of tardigrades, rotifers, and nematodes represent the apex of the soil foodweb (Freckman andVirginia 1997, Adams et al. 2006).Molecular analyses of microbial communities from mineral soils have revealed higher than expected diversity in comparison to non-polar soils, with an abundance of heterotrophic bacteria (;90% of total isolates) (Cary et al. 2010, Takacs-Vesbach et al. 2010).This suggests a carbon (energy) limitation to terrestrial microbiota, although extremes in pH and conductivity reported for this region can also exceed levels known to limit the abundance and distribution of microorganisms, here and elsewhere (Fierer andJackson 2006, Poage et al. 2008).  1 for an explanation of site labels.v www.esajournals.org

SAMPLE COLLECTION
We collected samples from 16 regional sites (8 each in Taylor and Wright Valleys; Fig. 1, Tables 1  and 2) chosen with the intent of including locations exhibiting the full range of soil primary production for these valleys.Regional site sepa-ration ranged from 0.3 km to 60 km.To capture fine-scale gradients of soil production, 3 local plots were chosen within a 25 m 2 area of each regional site on the basis of surface cryptogam cover.Thus, local plots (n ¼ 48) varied from dry, mineral soil without obvious surface production (e.g., Site 1) to intermittently saturated zones  v www.esajournals.orgalong stream, lake, and snowpack margins supporting dense cryptogamic mats composed primarily of either cyanobacteria or moss (e.g., Site 16).One location (Site 2) consisted of rockassociated cryptogams (e.g., hypoliths; Pointing et al. 2009) found locally within an otherwise dry, upland landscape.Samples were collected from local plots in December 2010 to characterize the following: (1) surface cryptogam chlorophyll a and (2) belowground soil and associated bacterial communities.A rectangular prism of cryptogamic mat of known surface area (,0.4 m 2 ) was separated from the mineral soil and collected into an opaque Nalgene bottle.The top 1 mm of mineral soil exposed by the removal of surface mat was additionally collected for chlorophyll a analysis, and in the absence of visual mat this top 1mm provided the sole estimate of producer biomass.For each local plot (n ¼ 48) this collection scheme was repeated in triplicate within a 2.5 m 2 area.Finally, ;500 g of bulk mineral soil was collected and pooled from beneath the areas sampled for cryptogams (at each local plot) to a depth of 5 cm.From this composite sample, ;10 g of soil was preserved in-field with a sucrose lysis buffer for nucleic acid stabilization (Mitchell and Takacs-Vesbach 2008).All samples were frozen within 12 hours of collection at À208C, with nucleic acids moved to À808C storage within 48 hours of collection.All samples were returned to the Blacksburg, VA campus of Virginia Tech for further analysis.

Soil productivity and biogeochemistry
Chlorophyll a concentrations were measured on composited surface mats and mineral soils (1 mm layer) as a proxy for soil productivity.
Although not a direct measure of production, this remains a quick and efficient technique for an index of potential productivity in soil and other habitats, as chlorophyll a is subject to rapid photochemical degradation and is therefore representative of living or only recently senesced tissue (Metting 1994).Chlorophyll was extracted from 2 mm sieved material using a dimethylsulfoxide (DMSO) extraction procedure under lowlight conditions adapted from Metting (1994) and Castle et al. (2011).Sample extractions (5 mL DMSO:1.5 g sieved material standard ratio) took place at 658C for 1 h with 20 s vortexing every 30 min.Samples were then centrifuged at 4000 rpm for 15 min.to pelletize particulates and a portion of the supernatant used for spectrophotometric analysis using a Shimadzu UV-1601 UV-VIS spectrophotometer (Shimadzu, Columbia, MD, USA).Absorbance at both 665 nm and 750 nm was recorded before and after dilute-acid degradation of chlorophyll a to account for phaeopigment content.All results were standardized to dry weight of starting material and thus expressed in w/w of chlorophyll a per mass of starting material.A second round of extraction performed on a subset of samples indicated an average first-round chlorophyll extraction efficiency of 85%, leaving many second round extracts too dilute for spectrophotometry.One extraction was used for all future samples.
Belowground biogeochemical parameters are reported per unit dry soil mass.A 1:2 and 1:5 soil/water slurry was used to assess soil pH and electrical conductivity, respectively, following standard procedures developed for this region (Nkem et al. 2006).Soil water content was determined gravimetrically by oven-drying for 48 hours at 1058C.Total soil organic carbon (SOC) and total nitrogen were estimated from ;300 mg of ground, dried, and acidified soil using a FlashEA 1112 NC Elemental Analyzer (CE Elantech, Lakewood, NJ, USA) (Barrett et al. 2009).A 1:5 soil/deionized water slurry was centrifuged and the supernatant analyzed for major soluble ions using standard ion chromatography methods (Thermo Scientific Dionex, Sunnyvale, CA, USA).Soil nitrate (NO 3 1À -N) concentrations were estimated using a 2 M KCl soil extraction and subsequent Lachat QuikChem 8500 Flow Injection Analyzer (Lachat Instruments, Loveland, CO, USA) assay of centrifuged extracts (QuikChem Methods 10-107-04-1-B).
Chloroform-labile carbon was used as an indication of soil microbial biomass and involved a 5-day fumigation of soil samples with gaseous chloroform (Cheng and Virginia 1993).Paired fumigated and non-fumigated samples were then extracted with a 0.5 M K 2 SO 4 solution and final extracts analyzed for total organic carbon using a OI Model 1010 Total Organic Carbon Analyzer (OI Analytical, College Station, TX, USA), where final chloroform-labile carbon was calculated as the difference between fumigated and nonfumigated total soil organic carbon.Soil inverte-brates were extracted using a modified sugar centrifugation method (Freckman and Virginia 1993), enumerated by microscopy into the three major taxonomic groups (rotifers, tardigrades, and nematodes), and later pooled into total invertebrate abundances.
Soil extracellular enzyme activity was assayed targeting a-and b-glucosidase to characterize major organic matter degrading enzymes.These hydrolytic enzymes are produced by microorganisms to initiate the decomposition of complex extracellular organic compounds into simple units (e.g., glucose) that are easily transported across the cell membrane.a-1,4 glycosidic bonds are common in starch and simple polysaccharides, while b-1,4 glycosidic bonds typify more structurally-complex compounds such as cellulose and chitin.The relative ratio of b/a-glucosidase enzyme activity, therefore, may serve as an indication of the relative complexity of soil organic matter pools (Sinsabaugh et al. 2010).Potential enzyme activity was measured using 0.5 g soil incubations with the labeled substrates 4-methylumbelliferyl(MUB)-a-D-glucopyranoside and 4-MUB-b-D-glucopyranoside in the presence of 50 mM NaHCO 3 buffer (pH ¼ 8.2) following the methods of Zeglin et al. (2009).Triplicate samples were incubated at room temperature on a platform shaker (250 rpm) for a minimum of 2 h and enzyme-induced fluorescence measured by excitation (360 nm) and emission (465 nm) using a Tecan SpectraFluor Plus plate reader (Tecan, Mannedorf, Zurich, Switzerland).In addition to sample incubations, control (buffer only), substrate (substrate þ buffer), and standard (standard þ buffer) references were analyzed to account for other sources of fluorescence.Final activity was normalized to sample soil organic carbon content and expressed as activity (nmol)Áh À1 Ág SOC À1 .

Bacterial communities
A terminal restriction fragment length polymorphism (TRFLP) procedure was chosen in order to provide an index of bacterial taxonomic richness and community structure in the soils of our local plots.TRFLP is a largely automated process suited for high sample through-put and remains particularly useful for tracking changes in microbial community structure over time and space (Schutte et al. 2008).The TRFLP method involves use of a fluorescent primer during DNA amplification, amplicon digestion with one or more restriction enzymes to produce DNA fragments of varied length, and fragment separation/quantification via capillary electrophoresis.Fragment relative abundance provides an estimate of community diversity and structure (Thies 2007).

Data analysis
All data analysis was restricted to a subset of 32 plots where DNA extraction was successful, and thus bacterial community information available.Plots were categorized into productivity classes based on levels of surface chlorophyll a using k-means non-hierarchical clustering with JMP statistical software, specifying three conservative a priori groups (Fig. 2).Univariate statistics (simple regressions, partial regressions, Spearman rank correlations) were performed on square-root transformed biogeochemical data using JMP statistical software to explore the relatedness among measured variables (JMP, Version 9, SAS Institute Inc., Cary, NC, USA).
MANOVA and multivariate tests/ordinations were performed using R statistical freeware (R Development Core Team).Mantel tests involved comparisons of three distance matrices (50000 permutations each): community (TRFLP) data transformed using a Bray-Curtis distance metric; decimal degree geographic data which was transformed using the earth.dist(fossil) R package (Vavrek 2011); and soil biogeochemistry data (scaled and centered) transformed using a euclidean distance metric.Principle components analysis (PCA) was performed of soil properties using a correlation distance metric and scores of the two primary eigenvectors were plotted.Nonmetric multidimensional scaling (nMDS) analysis was performed using bacterial community data and the Bray-Curtis distance metric with axes rotated to principle components.The final nMDS ordination represents a plot of site scores for the two primary axes.Canonical correspondence analysis (CCA) was used to ordinate weighted average, scaled (by eigenvalue) site scores under the constraints of multiple linear regression with environmental variables.Multi-response permutation procedure (MRPP) was used to assess differences among groups of response variables in ordination results.TRFLP results of fragment abundance was used for calculations of bacterial richness and Shannon-Weiner diversity, and were square-root transformed to reduce the distortional influence of high abundance taxa.

Soil productivity and biogeochemistry
Soil biogeochemical properties exhibited orders of magnitude variation across all 48 plots (Tables 1 and 2; Appendix: Table A1).For example, chlorophyll a concentrations exhibited v www.esajournals.orga gradient of soil productivity spanning more than three orders of magnitude (Fig. 2) and are comparable to those reported for hot desert soil biological crusts of North America (;5-10 lg chla/g soil) (Castle et al. 2011), although the range here is considerably greater.Organic carbon (mean ¼ 733.1 mg/kg dry soil) and total nitrogen (mean ¼ 92.0 mg/kg dry soil) concentrations also exhibited wide variation, again ranging over an order of magnitude, generating average molar C:N ratios of 9.2 6 1.7 SD.
Mean a-and b-glucosidase activities were 3590 and 4010 nmolÁh À1 Ág SOC À1 , respectively.These values are higher than previously reported for dry mineral soils in the McMurdo Dry Valleys (Zeglin et al. 2009) and are more comparable to those found in semi-arid deserts (Zeglin et al. 2007).Total invertebrate densities averaged 1200 individuals/kg dry soil, but ranged from none detected to over 6300 individuals/kg dry soil, similar to modal densities reported by others in similarly productive Antarctic soil environments (Barrett et al. 2006, Simmons et al. 2009).Microbial biomass averaged 25.5 mg/kg dry soil, within the range previously reported for dry valley soils (Barrett et al. 2006).
Correlations among these variables indicate a strong relationship between biological parameters such as chlorophyll concentration, soil organic carbon, total nitrogen, microbial biomass carbon, enzyme activity, invertebrate abundance, and TRFLP bacterial richness (Table 3).However, only chlorophyll was significant (using partial regression analysis) in predicting soil carbon concentrations (standard coefficient ¼ 0.68, all variance inflation factors ,1.8) in a model also considering pH, conductivity, and moisture.In turn, soil organic carbon was the only significant predictor of TRFLP bacterial richness in a model also including pH, conductivity, chlorophyll, microbial biomass carbon, and moisture (standard coefficient ¼ 0.84, all variance inflation factors ,5.8).
Because the influence of soil productivity on the subsurface environment was of specific interest for this study, we used k-means clustering to bin local plots into productivity classes to serve as predictors of underlying soil properties and microbial diversity in multivariate analyses.Fig. 2 depicts the results of this clustering, where 8 plots grouped below 2 lg chla/g dry material ('low productivity'), 15 plots between 2-35 lg chla/g dry material ('medium productivity'), and 9 plots above 35 lg chla/g dry material ('high productivity').With productivity class and regional site location as predictors, a multivariate ANOVA (MANOVA) was used to examine the amount of variability in soil properties which could be explained by these factors and their interaction.Results indicate that ad hoc soil productivity classes explain very significant levels of variation in organic carbon, total nitrogen, microbial biomass, and total invertebrates (all p 0.001), while enzyme activity ratios and bacterial (TRFLP) richness were also significantly constrained (p 0.01) (Table 4).In contrast, differences in geochemical parameters such as pH and conductivity were associated with regional variation among the sites in Taylor and Wright Valley.

Bacterial community diversity and structure
Soil DNA amplification was successful for a subset (32) of all 48 local plots, dictated primarily by levels of microbial biomass (multiple logistic regression, p ¼ 0.012).Final TRFLP results  1 and 2. *P 0.05; **P 0.01; ***P 0.001.v www.esajournals.orgindicate an average bacterial ribotype richness of 23.4 and an average Shannon Index (H') of 2.5 which is comparable, although somewhat lower, than those of Fierer and Jackson's (2006) global assessment of microbial biodiversity.A Mantel test was performed to correlate distance matrices of bacterial community similarity with both geography (spatial distance) and soil properties (environmental distance).The results of these tests indicate a much stronger correlation of communities with soil properties than with geography (Mantel's R ¼ 0.542, p , 0.0001; Mantel's R ¼ 0.210, p ¼ 0.006, respectively).When the influences of both spatial proximity and soil properties are controlled in these analyses (e.g., partial Mantel tests), the strength of correlation remained very high between bacterial communities and soil properties (Mantel's R ¼ 0.533, p , 0.0001), while the correlation between communities and geographic distance declined (Mantel's R ¼ 0.173, p ¼ 0.018).Finally, a distance decay plot using simple linear regression (data not shown) suggests that geographic separation explains only a small proportion of the variance in bacterial community similarity (r 2 ¼ 0.044, p , 0.0001).
Principle components analysis (PCA) was used to examine how soil habitats differ with respect to the primary soil properties of water content, pH, electrical conductivity, organic carbon, and total nitrogen (Fig. 3).PC1 (eigenvalue ¼ 1.53) and PC2 (eigenvalue ¼ 1.15) together were able to constrain 73.4% of the variation in soil properties, with little additional explanation from addition of further axes (eigenvalues , 1.0) (McCune and Grace 2002).Correlations of soil properties with  Notes: Abbreviations are as in Tables 1 and 2. *P 0.05; **P 0.01; ***P 0.001.
Multi-response permutation procedure (MRPP) of the ordination indicates a significant difference between points when grouped by the 3 ad hoc productivity classes (A ¼ 0.251, p ¼ 0.001).
A non-metric multidimensional scaling (nMDS) analysis was performed using TRFLP bacterial relative abundance data (Fig. 4).A stable solution was achieved within 20 iterations of the test (stress ¼ 0.178).Resulting axes were then correlated with soil environmental properties to assess which soil characteristics were associated with the spread of ordinated communities.Axis 1 appears most strongly related to electrical conductivity (r ¼ 0.87), while Axis 2 was negatively related to a number of properties associated with highly-productive plots, namely moisture (r ¼ À0.55) and soil organic carbon (r ¼ À0.48), along with a positive relationship to pH (r ¼ 0.51).As with PCA, MRPP found significance among plots when grouped by productivity class (A ¼ 0.042, p ¼ 0.001).
Finally, canonical correspondence analysis (CCA) was used to examine the direct relationship of bacterial communities to the soil properties of organic carbon, pH, and electrical conductivity, chosen based on their lack of multicollinearity (variance inflation factors , 1.1) and strong correlation with nMDS axes (Fig. 5).Five hundred test permutations yielded strong significance (p ¼ 0.002).The proportion of inertia (total variance) captured by constrained axes was low (18.8%), and eigenvalues for the first two axes were also small (0.543, 0.288, respectively).Correlation analysis indicates CCA1 strongly related to conductivity (r ¼ 0.94), while CCA2 is strongly related to pH (r ¼ À0.73) and soil organic carbon (r ¼ 0.62).

DISCUSSION
Heterogeneous biogeochemical conditions are a common characteristic of arid ecosystems (Aguiar andSala 1999, Wall andVirginia 1999).Such variation, particularly with respect to productivity and resource availability, has been widely used in macroecological research to examine the distribution and structure of both plant and animal communities and to test ecological theory explaining patterns in biodiver- v www.esajournals.orgsity (Abramsky and Rosenzweig 1984, Waide et al. 1999, Mittelbach et al. 2001).In the McMurdo Dry Valleys, productivity gradients associated with photosynthetic cryptogams are an important source of resources supporting soil organisms (Simmons et al. 2009) and contribute to the observed spatial patterns in diversity and community structure of bacterial communites described here.Surface chlorophyll associated with a productivity gradient spanning both local and regional scales was significantly correlated with subsurface soil properties including microbial biomass carbon, total nitrogen, and invertebrate abundance.Chlorophyll was also the best predictor of soil organic carbon, a vital resource in this energy-limited ecosystem.This demonstrates the linkage between above-and below-ground processes and soil properties in driving patterns of diversity, as has been noted for numerous temperate systems (Wardle et al. 2004).
Variation in biological properties such as total invertebrates, microbial biomass and diversity of bacterial communities was well explained by the stratification of study plots into productivity classes determined by clustering.Variation in geochemistry (e.g., pH, conductivity) was primarily related to regional differences in land-scape history and soil development (Table 4), as has been previously described by others (Bockheim 1997, Ugolini andBockheim 2008).Similar conclusions were drawn by Barrett et al. (2004) for biological communities in the Dry Valleys, where scale-dependent variation in soil properties (e.g., salinity, pH, and organic matter) significantly influenced spatial variation in invertebrate communities.
Bacterial community diversity was significantly correlated to soil organic carbon content suggesting that organic matter is a primary resource limitation to soil microbes in this system (Fig. 6).This relationship and other recent findings (Horner-Devine et al. 2003, Langenheder and Prosser 2008, Logue et al. 2012) illustrate signficant productivity-diversity relationships for microbial communities.Although the polynomial relationship we found does not conform to the unimodal curve often reported for diversity-productivity studies of macroorganisms (Waide et al. 1999, Mittelbach et al. 2001), the gradient we describe may represent a more restricted range of resource availability, insufficient to support highly competitive copiotrophic bacterial taxa (Fierer et al. 2007) and the competitive exclusion of other taxa as observed v www.esajournals.org in more productive ecosystems.Dry valleys zones of high productivity are also likely composed of increasingly diverse primary producer assemblages (e.g., mosses, chlorophytes, diatoms and cyanobacteria) (Cary et al. 2010), the activity of which may increase the structural complexity of the soil organic matter pool.Indeed, we observed a significant increase in the activity of complex carbon acquiring enzymes (increasing ratio of b/a glucosidase activity) in the most productive habitats.Resource (carbon substrate) richness may play an important role in facilitating the greater taxonomic diversity of microbial communities reported for such locations (e.g., Grayston et al. 1998, Orwin et al. 2006), but further research will be needed to determine the relative effects of resource quantity and quality (i.e., resource diversity) on microbial diversity and activity in such systems.
In contrast to diversity, variation in bacterial community structure appears to be more strongly influenced by geochemical properties, such as pH and conductivity, than resource availability.Ordination results (both nMDS and CCA) suggest that although significant differences exist in communities when grouped by ad hoc productivity classes, the association of multivariate axes with geochemistry (particularly electrical conductivity and pH) is generally stronger than the relationship with resources such as organic carbon or water (Figs. 4 and 5).Thus, although resource availability has a measurable influence on structure, the most divergent bacterial communities are primarily associated with extreme geochemical conditions (particularly when electrical conductivity exceeds 1000 lS/cm and pH is greater than 9.40).Such values may represent threshold levels beyond which specialized taxa predominate.Analogous thresholds in geochemistry have been shown to influence the presence and absence of invertebrate species in this region (Treonis et al. 1999, Poage et al. 2008).
Our results demonstrate that bacterial community dynamics are driven primarily by variation in soil properties (i.e., organic matter, pH and salinity), however a weaker ( yet still significant) effect of geographic distance is detectable using Mantel and partial Mantel tests.A linear regression of geographic distance versus community similarity indicates a weak, yet significant, pattern of distance decay (likely inflated by the high number of pairwise observations, n ¼ 496).Taken together these results suggest that although geographic processes may play a significant role in determining bacterial community structure in dry valley soils, for the locations observed here the influence of local soil conditions appears to be the primary driver.Because sampling was performed specifically to encompass extremes in both spatial distance and soil conditions, this may provide evidence for predominantly local, environmental controls over the distribution of dry valley microorganisms.
A bacterial diversity response to resource availability, and community structure response to environmental severity, mirrors effects demonstrated for both grassland plant ecosystems (Piper 1995) and experimental aquatic mesocosms (Chase 2010).Considering these examples, the minimal response in community structure to resource availability may be interpreted as taxonomic nestedness along a productivity gradient, where oligotrophic communities represent a diminished, tolerant subset of those taxa normally present under more eutrophic conditions.Similarly, the lack of response in community diversity along an environmental severity gradient may be interpreted as turnover of community members better adapted to geochemical extremes (saline/alkaline habitats) without modification of alpha richness.Still unknown are the temporal responses of dry valley bacterial communities to changing abiotic conditions caused by seasonal and other long-term dynamics, although it seems probable that alterations to the physicochemical environment will induce a response in bacterial community structure.Altogether this evidence suggests that bacterial communities can respond to abiotic controls in ways similar to macroorganisms, yet the response is context dependent, contingent on the nature of environmental change (e.g., altered resource availability or geochemical severity).The functionality of the soil microbiome may be subsequently dependent on whether bacterial diversity, structure, or both metrics respond to shifting environmental conditions.

SUPPLEMENTAL MATERIAL APPENDIX
Table A1.Regional sampling site locations and average (n ¼ 3) soil characteristics including a-glucosidase activity (activity/g organic carbon/hr), b-glucosidase activity (activity/g organic carbon/hr), nitrate-N (mg/kg dry soil), and sulfate (mg/kg dry soil) along with average soil characteristics of ad hoc productivity classes; standard deviation in parentheses.

Fig. 1 .
Fig. 1.Location of sixteen regional sampling sites in Wright and Taylor Valley of the McMurdo Dry Valleys, Antarctica.See Table1for an explanation of site labels.

Fig. 2 .
Fig. 2. Productivity (chlorophyll a) gradient observed for 48 local plots, log scale.Error bars represent standard deviation.Gradient divided by horizontal lines into three productivity classes created using k-means nonhierarchical clustering.Regional sites ranked by increasing maximum chlorophyll a content.Closed circles indicate locations where DNA could be extracted (32 of 48 plots).

Fig. 3 .
Fig. 3. PCA ordination of the soil properties moisture, soil organic carbon (SOC), total nitrogen (TN), electrical conductivity, and pH for 32 local plots.Biplot of soil properties, as correlated with major axes, is overlain.Labels indicate productivity classes determined via clustering of local plots by soil surface chlorophyll a concentrations.

Fig. 4 .
Fig. 4. nMDS ordination of square-root transformed relative abundance bacterial (TRFLP) data for 32 local plots.Labels indicate productivity classes determined via clustering of local plots by chlorophyll a concentrations.Biplot of soil properties moisture, soil organic carbon (SOC), total nitrogen (TN), electrical conductivity, and pH (as correlated with major axes) is overlain.

Fig. 5 .
Fig. 5. CCA ordination of 32 local plots using weighted average site scores.Labels signify productivity classes determined via clustering of local plots by chlorophyll a concentrations.Biplot of soil properties soil organic carbon (SOC), electrical conductivity, and pH is overlain to emphasize the relationship between site bacterial communities and environment.

Table 2
Note: N/A means data unavailable.

Table 3 .
Spearman rank correlations among soil variables.Notes: CHLA ¼ Chlorophyll a concentration (lg/g dry soil).All other abbreviations are as in Tables

Table 4 .
MANOVA results using productivity class (n ¼ 3) and regional site location (n ¼ 11) as predictors of common soil properties.