Patterns of bacterial biodiversity in the glacial meltwater streams of the McMurdo Dry Valleys, Antarctica

Microbial consortia dominate glacial meltwater streams from polar regions, including the McMurdo Dry Valleys (MDV), where they thrive under physiologically stressful conditions. In this study, we examined microbial mat types and sediments found in 12 hydrologically diverse streams to describe the community diversity and composition within and across sites. Sequencing of the 16S rRNA gene from 129 samples revealed ∼24 000 operational taxonomic units (<97% DNA similarity), making streams the most biodiverse habitat in the MDV. Principal coordinate analyses revealed significant but weak clustering by mat type across all streams (ANOSIM R-statistic = 0.28) but stronger clustering within streams (ANOSIM R-statistic from 0.28 to 0.94). Significant relationships (P < 0.05) were found between bacterial diversity and mat ash-free dry mass, suggesting that diversity is related to the hydrologic regimes of the various streams, which are predictive of mat biomass. However, correlations between stream chemistry and community members were weak, possibly reflecting the importance of internal processes and hydrologic conditions. Collectively, these results suggest that localized conditions dictate bacterial community composition of the same mat types and sediments from different streams, and while MDV streams are hotspots of biodiversity in an otherwise depauperate landscape, controls on community structure are complex and site specific.

cyanobacteria, microbial mats, diversity, community structure, polar region, 16S rRNA gene

INTRODUCTION

Microbial life dominates polar freshwater sediments, from biofilms to dense microbial mats, and includes filamentous cyanobacteria, chemotrophic bacteria and micro-eukaryotes (chlorophytes and diatoms). Microbial mats are one of the most biomass- and resource-rich habitats in high latitude regions due to characteristics that allow them to thrive under highly variable and physiologically stressful conditions (Vincent 2002; Vincent and Quesada 2012). These adaptations include highly efficient photon-harvesting systems that enable cyanobacteria, the dominant mat phototrophs, to grow in low light conditions occurring in fall through spring (Hawes and Schwarz 2001), and ancillary pigments that protect them from ultraviolet radiation stress during summer months (Roos and Vincent 1998). Furthermore, nutrient scavenging and internal nutrient cycling allow mats to flourish in oligotrophic conditions (Varin et al.2010), and exopolymers and cold shock proteins buffer the impacts of variable temperature, salinity (Hawes et al.1999) and desiccation (Varin et al.2012).

These characteristics have enabled mats and associated sediment communities to become the dominant biological features of glacial meltwater streams in the McMurdo Dry Valleys (MDV), Antarctica (Davey and Clarke 1992; McKnight et al.1998). For most of the year, microbes are frozen, desiccated and metabolically and reproductively dormant. Glacial meltwater occurring seasonally (Howard-Williams et al.1986) or after many years (McKnight et al.2007) can rapidly reactivate the metabolic and photosynthetic machinery of these mats (Vincent and Howard-Williams 1986). In the MDV, mat biomass, as well as diatom and bacterial community composition, is primarily controlled by stream geomorphology and hydrology (Stanish et al.2013), while the influence of water chemistry is modulated by internal biological and hyporheic interactions. In turn, mat and sediment microbes also influence instream conditions through biological uptake and processing such that nutrients to downstream lakes are reduced when streams contain significant mat biomass (Gooseff et al.2004; McKnight et al.2004).

Four microbial mat types, relatively distinct functional groups based on their hydrologic niches within the streambed (Howard-Williams et al.1986; Kohler et al.2015b), are recognized among MDV streams based on their color: orange, red, green and black. Microscopic observations suggest that orange and red mats are formed by filamentous Oscillatoria cyanobacteria (Vincent and Howard-Williams 1986; Alger et al.1997; Niyogi et al.1997), and are found in the main channel benthos of many streams. Green mats are often formed by chlorophyte genera and persist as filamentous streamers attached to rocks (Alger et al.1997). Lastly, black mats are limited to the wetted seeps and the visible expression of the underlying hyporheic zone along stream edges and are dominated by Nostoc (Alger et al.1997). Any combination of the four mat types may be present in a given stream (Kohler et al.2015b).

Previous comparisons of MDV stream mat types through morphological analyses reported orange and red mats to have greater taxonomic richness than green and black mats, and variation among samples collected from the same reach was as great as the variation among streams with widely different flow regimes (Alger et al.1997; McKnight et al.1998). However, limitations associated with morphological identifications (especially cyanobacteria; Broady 1982; Broady and Kibblewhite 1991; Alger et al.1997; McKnight et al.1998; Stanish, Nemergut and McKnight 2011) have probably led to an overall underestimate of mat biodiversity, limiting comparisons between sites and habitats. Furthermore, microbial communities living in the benthic sediments have been largely neglected (but see Zeglin et al.2011), despite their acknowledged role in nutrient transformations in Dry Valley streams (Gooseff et al.2004; McKnight et al.2004).

Modern molecular methods provide alternative techniques to investigate the community composition of stream microbes, and have been applied in a variety of polar habitats (Brambilla et al.2001; Jungblut et al.2005; Taton et al.2006; Fernández-Valiente et al.2007; Varin et al.2010; Larouche et al.2012; Peeters et al.2012; Zhang et al.2015). Most of these studies, however, have been constrained by a limited number of samples from few locations or were focused on a specific mat type. As MDV microbial communities provide critical biogeochemical links between glacial, aquatic and terrestrial landscapes, elucidating the currently underdescribed diversity, and taxonomic and functional identity of these organisms is essential for understanding MDV ecosystem function. Furthermore, as these streams are increasingly experiencing climate change-related disturbances including increased flow, vertical and lateral cutting, thermokarst formation, and altered nutrient and sediment regimens (Levy et al.2013; Fountain et al.2014; Gooseff et al.2016), a comprehensive assessment of the biodiversity from streams across this highly variable region is timely.

The purpose of this study was to describe the biodiversity and community structure of stream microbial mats and associated sediments using modern molecular methods. Based on previous findings in MDV streams of the hydrologic regulation of mat biomass and diatom community composition (Esposito et al.2006; Stanish, Nemergut and McKnight 2011; Kohler et al.2015b), our overall hypothesis was that bacterial community composition and biodiversity will vary greatly among streams and among mat types. We addressed three specific questions: (i) What are the identities of the major phototrophs, chemotrophic bacteria and eukaryotes from the four mat functional types and the sediments found in streams in the MDV region as determined by molecular techniques. (ii) How does microbial community composition differ among mat types and sediments both within and across streams. (iii) What abiotic factors are correlated with mat and sediment diversity and community composition. These results will aid in better understanding the general drivers of microbial mat diversity both in the MDV and elsewhere, interpreting monitoring efforts conducted in the region, and forecasting the future of the biological communities in the MDV in response to a potentially new ecological equilibrium.

METHODS

Sample collection

This study was conducted in January 2012 during the period of peak stream flow in the MDV, Victoria Land, Antarctica (77° 30^′ S, 163° 00^′ E). We sampled 12 streams that flow within the Taylor, Wright and Miers Valleys (Fig. 1), many of which have been the focus of hydrological, biological and chemical measurements made by the McMurdo Long Term Ecological Research (MCM LTER) program since 1990 (mcmlter.org). Stream sediment and four mat types, including black, green, orange and red mats, were collected when present (Table 1) along pre-existing microbial transects established by the MCM LTER in 1994 for six of the streams (see descriptions by Alger et al.1997 and Kohler et al.2015b) and at new transects for the remaining streams. While black and green (and some orange) mats from MDV streams are not vertically stratified in the manner commonly associated with benthic microbial mats, we have retained this term for consistency with previous MDV studies, as well as to provide a contrast between these cohesive structures and the loosely aggregated sediment communities.

Figure 1.

Stream sampling sites from Taylor, Wright and Miers Valleys. The presence of the mat types/sediment is indicated in the figure legend (black = black mat, green = green mat, orange = orange mat, red = red mat, and brown = sediment).

Table 1.

The number of mat and sediment samples successfully sequenced from each stream.

	LTER
Stream	transect	Black	Green	Orange	Red	Sediment
Adams		4	3	4	0	3
Aiken		4	0	4	0	3
Bohner	X	0	0	1	0	1
Canada	X	4	4	3	4	4
Commonwealth		4	0	4	0	3
Delta	X	0	4	0	0	4
Green	X	4	4	4	0	4
Lawson	X	4	4	4	4	1
McKay		0	0	2	0	2
Onyx at Lake Vanda		0	0	4	0	4
Onyx at Lower Wright		0	0	0	0	4
Von Guerard	X	4	4	3	0	4

The sampled streams spanned the range of lotic habitats present in the MDV, including streams with steady flow (Canada, Green, Commonwealth, Lawson and Adams), longer streams with intermittent flow that remain dry in some summers (Delta and Von Guerard) and shorter streams that drain glacial pools (McKay) or larger ponds (Aiken). Furthermore, two reaches were sampled on the Onyx River, the longest river in Antarctica, near its origin where it drains the large proglacial Lake Brownworth (along with several tributaries) and immediately before it empties into Lake Vanda. Samples were collected with EtOH-sterilized forceps (mat samples) or spatulas (sediment samples), transferred into sterile 15 ml conical tubes, preserved with equal volumes of sucrose lysis buffer (Mitchell and Takacs-Vesbach 2008) and immediately frozen at –20°C until further processing. Where available, four samples of each habitat (the four mat types and stream sediment) were collected from each site for analysis.

Biomass

For a subset of the mat samples (92), corresponding biomass-related parameters were measured from samples collected with a #13 brass cork borer (area of 227 mm²) using methods outlined in Kohler et al. (2015b). Briefly, mat samples were taken for chlorophyll-a (Chl-a) and ash-free dry mass (AFDM), and stored frozen and in the dark. AFDM samples were dried at 55ºC for 24 h, weighed, burned at 450°C for 4 h and re-weighed to determine mass lost on ignition (Steinman, Lamberti and Leavitt 1996). Chlorophyll-a was measured by extracting samples in 90% buffered acetone in the dark for 24 h, and were analyzed using a Turner Designs 10-AU field fluorometer (Turner Designs, Sunnyvale, California) (Welschmeyer 1994).

DNA extraction and amplification through PCR

Community DNA was extracted from all sediment and mat samples using the cetyltrimethylammonium bromide method described by Mitchell and Takacs-Vesbach (2008). Barcoded amplicon sequencing targeting the Bacteria (16S rRNA gene) and Eukarya (18S rRNA gene) was performed. Amplicon pyrosequencing of 16S rRNA genes was performed using universal bacterial primers 939F 5^′-TTG ACG GGG GCC CGC ACA AG-3^′ and 1492R 5^′-GTT TAC CTT GTT ACG ACT T-3^′ (Dowd et al.2008; Andreotti et al.2011). Approximately 50 ng of DNA per sample was amplified by a single step PCR to create 16S rRNA gene amplicons containing the Roche-specific sequencing adapters and a barcode (Hamady et al.2008) unique to each sample. PCR was performed in triplicate 25 μl reactions containing 1 μl 10 μM forward and reverse primer, 0.75 μl 25 mM MgCl₂, 2.5 μl 10X PCR buffer, 1U Platinum Taq Polymerase, 0.4 μl 12.5 mM DNTPs, 2 μl DNA template and sterile deionized water. The thermal cycler program consisted of 5 min at 94°C followed by 30 cycles of 30 s at 94°C, 30 s at 52°C, 1.5 min at 72°C and a final extension of 7 min at 72°C. PCR amplicons of appropriate size (∼550 bp) were purified using the MoBio Gel Purification Kit, quantified using a Nanodrop ND-2000c spectrophotometer, and individual samples were combined in equimolar concentrations. Sequencing library template was quantitated fluorometrically using a picogreen dye kit, assayed for quality and fragment length on an Agilent Bioanalyzer DNA 1000 chip before library preparation. Pyrosequencing was performed on a Roche 454 FLX instrument using Roche titanium reagents and titanium procedures in the Molecular Biology Facility at UNM Biology. All samples from this study were run on one two-region sequencing plate, distributed across the two regions, with no more than 96 samples total per region.

We also performed 18S rRNA gene sequencing of a subset (17) of the samples used for the bacterial analysis because significant portions of the 16S rRNA gene library included plastid/chloroplast sequence that was difficult to resolve taxonomically. These samples included two to six samples each of the four mat types (green, orange, red and black). Sequencing was performed at MR DNA (www.mrdnalab.com, Shallowater, TX, USA). Triplicate PCR was performed with barcoded eukaryotic primers euk1390F 5^′-GTACACACCGCCCGTC-3^′ and EukB-Rev 5^′-TGATCCTTCTGCAGGTTCACCTAC-3^′ (Amaral-Zettler et al.2009) following protocols given by the Earth microbiome project (http://www.earthmicrobiome.org/emp-standard-protocols/18s/) adapted for an Illumina MiSeq platform. Successful PCR reactions were pooled and cleaned with Ampure beads, combined in equimolar concentrations. The DNA library was prepared for sequencing using an Illumina TruSeq DNA kit and sequencing was performed using the MiSeq reagent kit v2 (300 cycle, 2×150 bp).

DNA sequence analysis

The 16S rRNA gene sequences were quality filtered, denoised, screened for PCR errors and chimera checked using default parameters in AmpliconNoise and Perseus (Quince et al.2011). The 18S rRNA gene sequences were quality filtered by removing bases with phred scores <20 or sequences less than 150 bp and then interleaved to join paired ends. Unique 16S and 18S rRNA gene sequences or operational taxonomic units (OTUs, 97% DNA identity) were identified using UCLUST de novo OTU picking (Edgar 2010) using the Quantitative Insights into Microbial Ecology (QIIME) pipeline (Caporaso et al.2010b). A representative sequence was chosen for each OTU, and aligned using the PyNAST aligner (Caporaso et al.2010a). Taxonomic assignments of unique 16S rRNA gene OTUs were made using the Greengenes core set (gg_13_5; DeSantis et al.2006) and by blasting against the NCBI database for the 18S rRNA gene data.

Alpha and beta diversity metrics were calculated for bacterial data only and were performed on randomly selected subsets of 750 DNA sequences per sample 1000 times to standardize for varying sequencing efforts across samples. Good's coverage, Chao1, Faith's Phylogenetic Diversity, Simpson (1D) and Shannon diversity metrics were calculated and statistical differences were determined using two-way analysis of variance (ANOVA). Linear regression was used to determine correlations between log-normalized measures of diversity and biomass measurements (AFDM, chl-a) for the subset of the mat samples with accompanying biomass data.

Bacterial community composition comparisons were made using unweighted and weighted UniFrac distance matrices (Lozupone and Knight 2005) on rarefied OTU tables and principal coordinate analysis (PCoA). UniFrac and taxonomic analyses were also performed on subsets of the dataset by filtering the original OTU tables according to taxonomic assignment to include only Cyanobacteria or phototrophs (+Cyanobacteria, rarefied to a depth of 200 sequences) or all OTUs except Cyanobacteria, i.e., chemotrophs (-Cyanobacteria, rarefied to a depth of 445 sequences) and by removing OTUs based on abundance (<10% of the sequences). ANOSIM in PRIMER was used to determine the statistical significance of differences in community structure by sample type (Clarke and Gorley 2006). Groups were designated as significantly different when the global test was significant (P < 0.05) or the pairwise test was significant (P < 0.10, due to the small sample size). To further resolve the community structure within each mat type, K-means clustering (MacQueen 1967) was performed in R (R Development Core Team 2011) with the ‘cluster’ package (Maechler et al.2012). K-means clustering is a non-hierarchical method analogous to the analysis of variance in reverse: cases are moved between clusters to minimize the within-cluster heterogeneity while maximizing between-cluster heterogeneity as determined by maximizing the silhouette width (Lattin, Carroll and Green 2003). The Random Forest analysis, a classification algorithm (Knights et al.2011), was performed in QIIME [with 10-fold cross-validation on a rarefied OTU (97% DNA identity) table, which was filtered to remove any OTUs with less than 10 sequences] to confirm the clustering patterns observed using ANOSIM and K-means analysis. The Random Forest analysis is a robust machine-learning technique for classification and regression that is appropriate for a wide diversity of data types, including microbial community data (Knights et al.2011). The effect of geographical distance on bacterial community composition was investigated by regressing the UniFrac distances against the linear distances between the sampling locations.

Environmental drivers of bacterial diversity and community composition

Pearson correlations were used to assess the relationships between community composition/richness and mean stream chemistry data for 5 years prior to sampling which is the approximate time required for mats to accumulate maximum biomass in MDV streams (Hawes and Howard-Williams 2013; Kohler et al.2015a). The chemistry samples were collected and analyzed according to Welch et al. (2010). To focus the analyses on the stream chemistry variables most likely to be important in controlling bacterial communities, we selected 5 of the 20 possible variables commonly found to be important in structuring bacterial communities and that had variance inflation factors under 2.00 (to avoid considering autocorrelated factors). Mean dissolved organic carbon (DOC), total dissolved solids (TDS), total nitrogen (N), soluble reactive phosphorous (SRP) and pH were compared to the 15 most abundant Cyanobacteria families and 30 most abundant bacterial families (excluding Cyanobacteria). Correlations between these environmental variables and Chao1 richness estimates were also calculated for orange mat and sediment samples as the other mat types occurred in too few of the streams for a robust analysis.

All raw sequence data from this study are available through the NCBI Sequence Read Archive. The individual raw sff files from this study were assigned the accession numbers SAMN02472266 through SAMN02472347 under Bioproject PRJNA228951.

RESULTS

Alpha diversity

Pyrosequencing of 129 samples from 12 streams resulted in 708 476 high-quality 16S rRNA gene sequence reads (mean = 5492, range = 1016 to 12 861). Using the 97% DNA sequence similarity criterion, 24 561 OTUs were identified and Good's coverage estimates ranged from 58% to 98% (mean = 87%). Chao1 richness estimates averaged 555 (range = 57–2293, Table 2). Richness was lowest in Adams, Aiken and Bohner streams and for green and orange mats, in general (P < 0.05, ANOVA). Phylogenetic diversity of the samples (Faith's Phylogenetic Diversity) varied among the streams and mat types similarly to Chao1 richness, with the exception that phylogenetic diversity was also significantly lower in McKay Stream and the black mats (P < 0.05, ANOVA). Simpson and Shannon diversity of the samples were less variable than Chao1 and Faith's phylogenetic diversity estimates. Simpson diversity did not vary significantly among the streams, but was significantly lower in the green mat type (P < 0.05, ANOVA). Shannon diversity was significantly lower in Bohner Stream and the orange and green mats (P < 0.05). Alpha diversity metrics by mat type are summarized in Table 2.

Table 2.

Average Chao1 diversity, Faith's Phylogenetic diversity, Simpson diversity and Shannon diversity, for bacterial samples from five mat types (colors) and stream sediment. Each value represents the average of 1000 iterations, with the standard deviation in parentheses.

Mat Color	n	Chao1	PD	Simpson	Shannon
Black	28	657 (397)	205 (86)	0.85 (0.20)	5.15 (1.57)
Green	23	304 (152)	104 (64)	0.70 (0.23)	3.41 (1.58)
Orange	33	437 (258)	152 (70)	0.72 (0.24)	3.95 (1.64)
Red	8	793 (359)	255 (58)	0.93 (0.03)	5.82 (0.49)
Sediment	37	691 (368)	271 (93)	0.91 (0.13)	6.06 (1.45)

Taxonomic composition

The 24 561 16S rRNA gene OTUs detected in the samples comprised over 40 Bacterial phyla. Cyanobacteria (36%), Bacteroidetes (29%), Proteobacteria (15%), Acidobacteria (7%) and Firmicutes (4%) dominated the overall taxonomic abundances. Unclassified Bacteria (3%), Planctomycetes (2%), Verrucomicrobia (2%), Thermi (1%), Armatimonadetes (1%) and Actinobacteria (>1%) were present in lower abundances. The most diverse phylum was Proteobacteria with 6131 OTUs. The phylum level distribution of OTUs was similar between all K-means clusters and mat types (Fig. 2A, see Fig. S1 for the taxonomic distribution of OTUs for individual samples.). Sediment samples had the lowest percentage of cyanobacterial OTUs (26%) and had the most even distribution of the top four most abundant phyla as compared to the mat samples (Fig. 2A).

Figure 2.

Stacked bar graphs of the OTUs derived from 16S rRNA gene sequences at the phylum level for all bacterial sequences (A), the order level for chemotrophic bacteria (B), and the genus level for Cyanobacteria/chloroplasts (C). The samples are averaged by the identified K-means clusters and are organized by mat/sediment type (black = black mat, green = green mat, orange = orange mat, red = red mat, and brown = sediment). See Fig. S1 for the taxonomic distribution of OTUs for individual samples.

Filtering to produce a non-cyanobacterial dataset reduced the dataset to 125 samples with greater than 445 sequences (mean = 3620, range = 446–10 756). At the taxonomic level of order, chemotrophs were dominated by Sphingobacteriales (Bacteroidetes, 43%), Burkholderiales (Betaproteobacteria, 9%) and Chloroacidobacteria (Acidobacteria, 8%) (Fig. 2B). Some members of the Chloracidobacteria are photosynthetic (Bryant et al.2007); however, those detected in this study share only 82% DNA similarity (over 410 bp) with known phototrophs. Comparison of the MDV stream 16S rRNA gene sequences classified as Chloroacidobacteria to the NCBI database by blastn indicates that they most closely match other uncultured Acidobacteria from other cold regions, especially soils associated with glacier forefields (96 to 100% DNA similarity).

Following filtering to produce a Cyanobacteria/chloroplast-only dataset, 121 samples remained containing 200 or greater sequence reads (mean = 2100, range = 206–7674). Each of the microbial mat types featured high abundances of several Cyanobacterial genera (Fig. 2C). However, these genera were not evenly distributed across all streams for a given mat type as seen in the variation between the averaged data for the K-means clusters (Fig. 2C). The black mats from Adams, Aiken and Von Guerard streams were dominated by Nostoc (Cluster Blk-03, OTUs 14336/3363, mean abundance of 44 ± 34% across all black mats), while the black mats from Canada (Cluster Blk-01), Commonwealth (Cluster Blk-01), Green Creek (Cluster Blk-02) and Lawson (Cluster Blk-04), all of which have a steady flow regime, also contained high proportions of Pseudanabaenaceae (OTUs 7126/3669/11192, mean abundance of 21 ± 23% across all black mats).

Synechococcophycideae was widely distributed across green mats (OTU 18488, mean abundance of 30 ± 20% across all green mats) with large proportions of chloroplasts found in mats from Green Creek and Lawson (Cluster Grn-02, OTU 7904, mean abundance of 22 ± 34% across all green mats). Orange mats from Adams, Aiken and McKay were dominated by a member of the Microcoleus genus (Cluster Orng-01, OTUs 4166/23411, mean abundance of 35 ± 39% across all orange mats). This organism has historically been classified as part of the Phormidium genus; however, recent phylogenetic analysis suggests that a reassignment to Microcoleus is likely appropriate (Strunecký, Elster and Komárek 2010). Synechococcophycideae was dominant in orange mats from Canada, Lawson and Onyx River at Vanda (Cluster Orng-04, OTU 18488, mean abundance of 34 ± 37% across all orange mats). The dominant cyanobacterial genus in red mats was Leptolyngbya (OTUs 18669/24559, 44 ± 20%). The cyanobacterial/chloroplast composition of sediment samples was highly variable; however, based on plastid sequences, there was a high proportion of Stramenopiles in sediments from Aiken, Commonwealth, Delta, Green Creek and Von Guerard (Clusters Sed-02 and 05, OTUs 10864/3967, mean abundance of 24 ± 30% across all sediment samples).

To resolve the chloroplast/plastid sequences observed in mats, we performed an analysis of 18S rRNA gene amplicons for a subset of 17 samples: 3866 OTUs were detected (97% DNA identity) including sequences from 12 phyla (Fig. 3A). Eukaryotic OTUs were dominated by the Rotifera (25%), Tardigrada (22%), Ascomycota (14%) and unidentified Eukaryota (13%). Bryozoa, Basidiomycota, Chlorophyta, Nematoda and Bacillariophyta contributed 2%–8% of the 18S rRNA gene sequences, and Streptophyta, Chytridiomycota and Glomeromycota contributed less than 1% each. No clear pattern of eukaryotic community composition was apparent among the four mat types (Fig. 3A) or by considering phyla that included algae (Bacillariophyta, Chlorophyta and Streptophyta; Fig. 3B). Although 47 algal OTUs were detected, 89% of the sequences were classified among only nine genera, including three diatom genera (Bacillariophyta), five genera from the Chlorophyta and one moss (Streptophyta) (Fig. 3B).

Figure 3.

Stacked bar graphs of the OTUs derived from 18S rRNA gene sequences at the phylum level for all eukaryotes (A) and at the genus level for photosynthetic eukaryotes (B). The samples are organized by mat/sediment type (black = black mat, green = green mat, orange = orange mat, red = red mat, and brown = sediment).

Beta diversity

We observed a weak, but significant association among the samples by stream and habitat (mat/sediment) type (P < 0.001, weighted UniFrac distances, ANOSIM r-statistics of 0.29 and 0.28, respectively, Fig. 4). Similarly, we observed a weak stream and habitat type association in ordinations when the data were filtered to include only cyanobacterial OTUs (data not shown). The strongest and most significant (r-statistic > 0.4, P < 0.05) clustering of the samples was observed by habitat type within stream sites (Fig. 5). Clustering was not strengthened by removing OTUs that contributed less than 10% of dataset sequences. Clustering patterns varied among the streams. For example, all habitat types were statistically distinct in Commonwealth and Von Guerard streams, but in Green Creek and Canada Stream, which have the most steady flow regime, significant overlap occurred between habitats. Aiken Creek and the Onyx River at Vanda site were the only two sites where significant clustering of at least one habitat type was not observed. K-means analysis was also used to detect clustering within mat types with 18 clusters detected: 4 clusters in black mat samples (found in 7 streams), 2 clusters in green mat samples (found in 6 streams), 5 clusters in orange mat samples (found in 9 streams), 2 clusters in red mat samples (found in 2 streams) and 6 clusters in sediment samples (found in 12 streams). The results of the Random Forest analysis confirmed that habitat and stream were both significant, but weak predictors of clustering, while mat type within stream was the strongest predictor (Table 3). For example, ratios of random error to model error were 4.6 for habitat and stream, but 7.6 for habitat by stream (ratios greater than 2 are expected for factors that can be accurately predicted; Knights et al.2011). The correlation between UniFrac and geographical distances was not significant (P > 0.05) for either all of the samples together (R² = 0.007) or when samples from the different habitat types were considered individually (R² = 0.0001 to 0.1).

Figure 4.

Principal coordinate analysis (PCoA) of UniFrac weighted distance matrices created from rarefied OTU tables containing all phototrophic and chemotrophic OTUs for all mat samples color coded by lake basin (A) and mat type (B).

Figure 5.

PCoA of UniFrac weighted distance matrices created from rarefied OTU tables for mat and sediment samples within individual streams color coded by mat/sediment type (black = black mat, green = green mat, orange = orange mat, red = red mat, and brown = sediment). Significant ANOSIM clusters are designated by dashed (P < 0.10) or solid (P < 0.05) circles.

Table 3.

Results from Random Forest classifiers.

Factors	Ratioa	Predictive accuracy (%)
Basin	1.8	78
Stream	4.7	82
Habitat	4.7	85
Habitat within stream	7.6	88

a

Ratio of random error to model error.

Correlations between bacterial populations and biomass and geochemical parameters

We observed significant (P < 0.001, R² = 0.47) positive non-linear correlations between AFDM and Unweighted UniFrac PCo1 for the subset of 92 samples for which a suite of biomass-related parameters was available. This relationship was verified with a Mantel test (999 permutations, R = 0.27) using the Bray–Curtis distance matrix of AFDM distances and the Unweighted UniFrac distance matrix. We observed a significant positive (P < 0.001) but weaker (R² = 0.36) non-linear correlation between Chl-a and Unweighted UniFrac PCo1; however, the Mantel test was not significant (P > 0.05). Correlations between PctN, PctC and PCo1 and their associated Mantel tests were not significant (P > 0.05). Log-normalized AFDM was also significantly correlated with several measures of diversity (also log normalized) including Chao1 (R² = 0.30), and Faith's Phylogenetic Diversity (R² = 0.41, Fig. 6).

Figure 6.

Linear regressions for AFDM and two diversity measures, Chao1 which is a non-parametric lower-bound richness estimator, and phylogenetic diversity (PD) calculated using the Faith Phylogenetic Diversity Metric.

Correlations between stream chemical parameters (Table 4) and the Chao1 diversity estimates for orange mats and sediments revealed a single strong negative correlation between Chao1 diversity and TDS (R = 0.71). Analysis for correlations between stream chemistry and specific bacterial family abundance revealed limited associations along pH or solute gradients. Cyanobacterial abundances, in particular, Nostoc and several oscillatorian groups including Phormidium, were positively correlated with DOC, TDS and total nitrogen levels (Fig. 7A). Pseudanabaena were positively correlated with SRP, whereas Chamaesiphon were negatively correlated. Correlation patterns among other Bacterial phyla (excluding Cyanobacteria) were less clear and stream chemistry did not significantly influence group abundances (Fig. 7B). The strongest correlations were between increasing Carnobacterium abundances with DOC and decreasing abundances of Thermi with pH.

Figure 7.

Heatmap of the correlations between stream chemistry parameters (5 of the 20 possible variables with variance inflation factors under 2.00) and the 30 most abundant Bacteria families excluding Cyanobacteria/chloroplasts (A) and the 15 most abundant Cyanobacteria/chloroplast families (B). Clustering was performed using the ‘dendrogram’ command in the heatmap.2 package in R. Abbreviations are as follows: TotalN, total nitrogen; SRP, soluble reactive phosphorus; TDS, total dissolved solids; DOC, dissolved organic carbon.

Table 4.

Average stream chemistry data collected from the MCMLTER database (mcmlter.org) for the 2007/08-2011/12 field seasons. Total N (nitrogen) includes NO₃ (nitrate), NO₂ (nitrite) and NH₄ (ammonium). TDS (total dissolved solids) is calculated as the sum of all major ions. SRP is an abbreviation for soluble reactive phosphorus and DOC is an abbreviation for dissolved organic carbon.

Stream	Total N	SRP	TDS	pH	DOC
	(uM)	(uM)	(mg l−1)		(mM)
Adams Stream	11.92	0.205	38.91	6.93	0.01
Aiken Creek	10.58	0.675	45.44	7.86	0.065
Bohner Stream	16.6	0.365	70.42	7.04	0.075
Canada Stream	1.83	0.13	9.03	7.31	0.031
Commonwealth	9.06	0.523	20.42	7.39	0.027
Delta Stream	1.85	0.183	60.17	7.77	0.03
Green Creek	3.57	0.242	11.86	7.39	0.047
Lawson Creek	23.55	0.18	20.85	7.25	0.031
McKay Creek	3.37	0.57	11.75	7.34	0.025
Onyx River (Lower Wright)	1.6	0.1	13.22	7.43	0.039
Onyx River (Vanda)	4	0.123	24.03	7.57	0.027
Von Guerard Stream	11.06	0.484	38.82	7.61	0.047

DISCUSSION

In this study, we examined the diversity, composition and distribution of bacteria and eukaryotes inhabiting four distinct microbial mat types (Alger et al.1997; McKnight et al.1998; Kohler et al.2015b) and corresponding benthic sediment of 12 hydrologically diverse MDV streams across three valleys to further understand differences in diversity and community composition and their respective controls. This work builds on numerous previous studies of MDV stream mats by exploiting recently available DNA sequencing methods that enable the interrogation of a large number of samples. We hypothesized that given the range of hydrologic regimes and stream characteristics, MDV stream bacterial biodiversity would be extensive, a prediction that was observed. However, the controls on community structure were complex and site specific. While outside of the focus of this research, hydrologic regimes at both spatial (riffle vs pool, stream vs river) and temporal (duration, magnitude, frequency, intermittency) scales are likely important for structuring different communities (Besemer et al.2007, 2009), and should be considered in the future.

Alpha diversity

Overall diversity patterns: The bacterial diversity detected here included over 24 000 OTUs, which is much greater than the cumulative diversity (4128 OTUs) from a recent MDV soil survey that included 88 samples and a similar geographical scale (Van Horn et al.2013). Additionally, the average alpha diversity of even the least diverse stream habitat type (green mats, mean Chao1 value of 304 OTUs) was equivalent to that of the most phylotype-rich soils in the MDV (Taylor Middle site, mean Chao1 value of 281 OTUs), while the most diverse stream habitats (red mat and sediment, mean Chao1 values of 793 and 691 OTUs, respectively) were two to three times as diverse as high diversity MDV soil habitats (Lee et al.2012; Van Horn et al.2013). Similar levels of biodiversity were previously reported in a study of microbial mats from five MDV streams (Stanish et al.2013), confirming that streams are biodiversity hotspots in the MDV.

Microbial mats and associated sediments from a wide range of environments contain greater diversity than surrounding habitats (Ley et al.2006; Bolhuis and Stal 2011; Harris et al.2013). Several underlying mechanisms interact to promote this diversity including the high productivity and niche complexity of mat habitats. Microbial mats, particularly in polar environments (Vincent 2002), contain large quantities of energy rich biomass and serve as islands of fertility in an oligotrophic landscape. In the MDV, this biomass is correlated with diversity as seen by the significant positive relationship between AFDM and several alpha diversity metrics. These observations are theoretically supported by the potential energy hypothesis, which proposes that more productive environments contain more individuals, which in turn, may support a greater number of ecologically stable populations (Gaston 2000). The well-developed biomass in cyanobacterial mats also leads to vertical complexity with a wide range of available light, strong chemical gradients including pH, dissolved oxygen, hydrogen sulfide and biologically produced pigments (Jorgensen, Revsbech and Cohen 1983; Ley et al.2006; Hawes, Giles and Doran 2014). Similarly, stream sediments and hyporheic zones, both generally and in polar environments, have highly heterogeneously distributed organic matter deposits, upwelling and downwelling regions, and redox gradients (Boulton et al.1998; Gooseff et al.2004). In stream mats and sediments, these varying conditions provide niche diversity within individual samples, promoting the observed high levels of diversity as predicted by niche theory (Chesson 2000; Levine and HilleRisLambers 2009). Similarly, for planktonic communities, metabolic partitioning along the light spectrum results in high niche differentiation (Stomp et al.2004), and stratification, microniches and complex mutualistic relationships support high diversity in non-cyanobacterial biofilm communities (Costerton et al.1995; Davey and O'Toole 2000).

Comparison of diversity by habitat (mat/sediment) type: This study is the first to sequence samples from each of the four distinct MDV mat types characterized by early studies in this region (Alger et al.1997; McKnight et al.1998) in addition to stream sediments. In agreement with these studies, we found that diversity varied by mat type, which suggests a relationship between diversity and mat characteristics such as color, morphology and microhabitat distribution. Green mats, which had the lowest diversity (significantly lower diversity than all but orange mats, mean of 304 OTUs), are found attached to rocks in the main channel and are composed of loose filaments with minimal vertical structure as compared to other mat types. Low organic matter accretion minimizes biological stratification across light and chemical gradients, potentially limiting diversity. Another limiting factor may be that these mats are subject to scour during high flows and take several years to reestablish (Kohler et al.2015b).

Orange mats are thicker and more vertically structured (Vincent et al.1993; Alger et al.1997; McKnight et al.1998), allowing the development of an organic matter rich, stratified environment with higher bacterial diversity (mean of 437 OTUs) than green mats. Black mats, which lack sophisticated stratification and generally contain a single layer dominated by Nostoc (Vincent et al.1993), interestingly had higher richness (mean Chao1 of 657 OTUs) than orange and green mats. Black mats contain abundant organic carbon and nitrogen resources (due to mucilage and N fixation by Nostoc and Pseudanabaena), which likely subsidize a diverse chemotrophic community. Alternatively, sediment and the associated microbes may have been collected with the black mats, increasing the diversity of these samples. The mechanisms underlying this unexpectedly high diversity in a structurally simple mat require further study. The highly diverse red mats (mean Chao1 of 793 OTUs) grow in thick, dense, rubbery-textured formations making this the most well developed, vertically structured mat type found in the MDV. Interestingly, red mats are both among the most stable and least common community types present in the MDV, with little change in established communities from year to year over the last two decades, and occur in streams with dependable annual streamflow (Kohler et al.2015b). These features likely promote increased niche and energy-related diversity.

This was one of the first studies in the MDV to examine the bacterial community composition of stream sediments (but see Zeglin et al.2011), which, in spite of an absence of structured mat communities, were one of the most diverse habitats. This high diversity may be derived from sloughed cells detached from mats (Cullis, Stanish and McKnight 2014) and deposited in low velocity areas, representing a reservoir of organisms from the relatively distinct habitats found within a given stream. Alternatively, the biogeochemical gradients within the MDV sediments (Gooseff et al.2002, 2003) provide multiple niches expected to promote bacterial diversity, as well as nutrient upwelling zones due to the ‘swiss cheese’ model of the hyporheic zone (Cozzetto et al.2013).

Taxonomy

The dominant organisms we found correspond with other studies of stream mat communities in the MDV, supporting the idea of a metabolically and species diverse microbial ecosystem containing several dominant species of Cyanobacteria and a wide diversity of Bacteria (Alger et al.1997; Michaud, Šabacká and Priscu 2012; Stanish et al.2013). Of the significant diversity described in this study, approximately 35% of the total sequences were identified as Cyanobacteria and 65% were from other bacterial phyla. Similar results were reported by Varin et al. (2012) in a study of polar cyanobacterial mat metagenomes, which found that non-cyanobacterial sequences made up a greater portion of the protein encoding genes than did Cyanobacteria in mats from Arctic and Antarctic ice shelves. A dominance by non-cyanobacterial sequences has also been observed in mats from a wide variety of habitats including unlaminated thrombolites (Mobberley, Ortega and Foster 2012), coastal beaches (Bolhuis and Stal 2011) and freshwater lakes (Sorokovikova et al.2013). While the dominance of aquatic microbial mats by non-cyanobacterial microorganisms may be at least partially attributed to unquantifiable experimental error, such as PCR primer bias, the presence of such a rich diversity of bacteria in general points to the existence of a complex consortium of chemotrophic organisms that is often overlooked because Cyanobacteria and algae are the most conspicuous mat members.

The community composition of the MDV mats can be interpreted at various levels including the broad phylum level for the entire community or the species level for some groups of organisms. At the phylum level, regardless of mat type, we found a core community dominated by Cyanobacteria, Bacteroidetes and Proteobacteria. These are the same major taxonomic groups found by other studies of Cyanobacterial mats from the MDV (Stanish et al.2013), other Antarctic and Arctic habitats (Varin et al.2012), freshwater lakes (Sorokovikova et al.2013), marine costal environments (Allen et al.2009; Bolhuis and Stal 2011; Mobberley, Ortega and Foster 2012) and saline ponds (Ley et al.2006; Lindemann et al.2013). Thus, as with other habitat types such as soils (Fierer et al.2009), the conditions found in cyanobacterial mats select for a predictable subset of phyla. At the genus level for the Cyanobacteria, the most abundant phylum, we observed many members common to other polar mats (Jungblut, Lovejoy and Vincent 2010; Kleinteich et al.2012, 2014; Lionard et al.2012). However, the absence of consistent taxonomic designations by mat type and the presence of numerous groupings therein suggest that these are indeed functional, as opposed to taxonomic, groupings.

The taxonomic distribution of the sediment samples revealed several interesting patterns. For both the broad phylum level and the chemotrophic genus level data, the high abundance of Acidobacteria and other unidentified bacteria and the low abundance of Cyanobacteria in sediments as compared to mat samples are consistent with findings from a recent review paper that compared the bacterial taxonomy from a variety of stream habitats (Zeglin 2015). The high frequency of Acidobacteria suggests that MDV stream sediments share taxonomic similarities to soils, as Acidobacteria are one of the most dominant soil phyla in this region (Van Horn et al.2013). The elevated frequency of unidentified bacteria in stream sediments has been attributed to the high degree of physical, chemical and redox gradient heterogeneity which may increase the abundance of narrow or unexplored phyla (Zeglin 2015).

We also performed a preliminary investigation of the eukaryotic organisms associated with these mats to fully describe mat diversity and to definitively identify the major phototrophs. For example, the dominant primary producer in green mats has been previously identified as Prasiola (Trebouxiophyceae) based on accounts which utilized morphological identification techniques (Alger et al.1997; McKnight et al.1998). However, sequences identified as members of the Prasiolales did not comprise more than 4% of any of the samples. Rather, the green algae associated with the mats were largely assigned to the Chlorococcum and the Chlorella. Underrepresentation of Prasiola in current sequence databases or PCR primer bias may contribute to the absence of Prasiola in our dataset, and it will be necessary to perform more targeted probing to approach more quantitative estimation of the importance of Prasiola in these mats. In addition, only three diatom genera were identified in these samples, which also may reflect the limitations of the PCR primers used. Microscopic analysis of both cleaned silica frustules and live samples has found that the diatom flora in streams is diverse, with 15–25 taxa found in each mat sample, and roughly 50 taxa from over 20 genera are commonly found in MDV streams (Esposito et al.2006, 2008; Stanish, Nemergut and McKnight 2011). Given these initial results, we suspect that the overall eukaryotic biodiversity in MDV stream mats is high and spatially variable, similar to the results from the Bacteria sequences.

Beta diversity

The breadth of sampling we performed allowed us to investigate within and across-site variation in bacterial community composition. One of the first goals of this study was to determine if different mat types from the same stream contained discernably different communities. The significant and strong clustering within most streams by mat type/sediment (Fig. 5) confirms that at a given stream site the different habitats contain distinct communities. This is consistent with results from other studies. For example, Mobberley, Ortega and Foster (2012) found that closely located unlaminated thrombolite mats of different colors from an intertidal zone had distinct bacterial communities, and Bolhuis and Stal (2011) found significant differences between bacterial communities from three adjacent coastal mat types spanning a salinity gradient. Thus, it appears that visually different habitat types from similar locations contain unique communities.

Another goal of this spatially comprehensive study was to determine if communities from sediment or one mat type were the same across streams. The low ANOSIM r-statistic values when samples were clustered by mat suggest that this is not the case: a black, green, orange, red mat or sediment sample from one MDV stream does not necessarily contain the same community as the same colored, functionally similar, mat or sediment from a different stream. Similar results were reported for orange mat communities from five MDV streams from three different valleys in which significant differences were found between sites (Stanish et al.2013). Additionally, in the extensively studied Guerrero Negro mats, samples from a 1 km transect where the mats visually looked similar clustered by sampling location, with the largest differences occurring at spatial scales greater than ∼10 m (Dillon et al.2009). Results from these studies are consistent with our findings in the MDV of weak grouping by habitat type across all sampling locations. This suggests that, as with soil communities in the MDV (Van Horn et al.2013), localized conditions create what have been termed ‘contextual effects’. These factors dictate the variability in the composition of the bacterial communities of the same habitat types from different streams. Stream-specific physical (House, McKnight and Von Guerard 1995), chemical (Welch et al.2010), discharge (Conovitz et al.1998) and temperature conditions all likely play a role in creating these contextual effects. Interestingly, a recent study of benthic mat cyanobacteria in MDV lakes found minimal variation in community structure between three lakes from different basins, suggesting that organisms inhabiting these relatively more stable environments are both widely distributed and have broad habitat tolerances (Zhang et al.2015).

The non-significant correlation between UniFrac and geographical distances for a given habitat type suggests that dispersal is not a major determinant of community composition and is thus not the cause of the observed differences between mats of the same functional group from different streams. For example, one of the K-means groupings of cyanobacteria from orange mats contained samples from Miers, Fryxell and Hoare basins, indicating that geographical barriers do not impose strong controls on these communities. Additionally, some of the largest differences in community composition occurred at the within region scale between streams in close geographical proximity. For example, Lawson Creek and Bohner Stream, which both flow into Lake Bonney, had drastically different bacterial communities; Bohner Stream is 60% dominated by Firmicutes, while Lawson Creek features less than 1% Firmicutes. The greater importance of environmental selection versus dispersal processes in structuring microbial communities has been described in many studies and has been summarized in recent reviews of general microbial ecology (Hanson et al.2012) and the ecology of stream microbes (Besemer 2015).

Relationships to geochemistry and the potential importance of hydrologic regime

The correlation analyses we performed between stream chemistry parameters and both diversity estimates and the abundance of specific bacterial families were a first step to test whether environmental filtering drives diversity patterns and the clustering of geographically disparate mat and sediment samples. The single significant negative correlation between orange mat/sediment diversity and TDS was similar to a negative correlation observed between electrical conductivity and biofilm diversity in near-glacier streams (Wilhelm et al.2013). However, the mechanisms underlying this relationship are unclear. In spite of the well-recognized roles of temperature, pH, nutrients, organic matter composition and flow regimes in controlling biofilm diversity and community composition (Besemer 2015; Zeglin 2015), we observed limited relationships between the dominant cyanobacterial/chloroplast and chemotrophic taxonomic groups with stream chemistry.

While the streams studied span the full range of habitats from three valleys, some of these streams lack sufficient discharge data to compare the mat communities with the hydrologic regime. This approach has been used previously in studies of mat biomass and diatom communities (e.g. Esposito et al.2006; Stanish, Nemergut and McKnight 2011; Kohler et al.2015b) in which hydrologic regime was found to be a major factor in structuring communities. The hydrologic regime can vary greatly among MDV streams, with some experiencing steady flow every season for most of the summer, whereas others may have no, limited, or intermittent flow (Stanish et al.2012). Streams with predictable, steady flow typically have abundant, diverse mats, whereas mat abundance is typically sparse in streams with less regular flow and in reaches with unstable streambed substrata (Kohler et al.2015b). Additionally, depending on the stream geomorphology, flow events may either scour mat biomass during daily flood pulses that occur with the changing sun angle, and during major flood events, or wet side channels that support additional biomass (Cullis, Stanish and McKnight 2014; Kohler et al.2015b). Thus, future studies are necessary to determine if hydrologic regime drives between-stream differences in bacterial community composition of samples from the same functional group, as is seen with mat biomass and mat associated diatom communities.

CONCLUSIONS AND FUTURE DIRECTIONS

Microbial mats and sediments from MDV streams are the most biodiverse assemblages currently described in this region. For mats, this diversity is well correlated with biomass variables suggesting that niche partitioning, within mat nutrient cycling and available energy play a larger role in driving diversity patterns than ambient nutrient concentration and stream chemistry. Despite sharing similar macromorphology and habitat, significant between-site variation exists within mat types and sediments, particularly for the phototrophic communities. These results support the emerging importance of contextual effects in structuring MDV microbial communities: unique combinations of variables at individual sites differentially impact the bacterial community structure, and is an important consideration for investigators moving forward into the coming decades of change.

SUPPLEMENTARY DATA

Supplementary Data.

The authors thank Rae Spain, the 2011/2012 MDV Stream Team, and PHI Inc. for logistical and field support, and Jeremy Dyke and Kelli Feeser for laboratory assistance.

FUNDING

This work was supported by the National Science Foundation [1115245 to C.D.T.V., 1142102 to C.D.T.V. and D.J.V.H., and 1245991 to D.J.V.H.]. The DNA sequencing performed at the UNM Molecular Biology Facility was supported by the National Institute of General medical Sciences of the National Institutes of Health under Award Number P30GM110907. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of interest. None declared.