Introduction

Over the last 25 years, cultivation-independent small subunit ribosomal rRNA (16S rRNA) gene surveys of naturally occurring microbial communities have uncovered a plethora of previously anonymous microbial diversity (Rappé and Giovannoni, 2003; Sogin et al., 2006; Pace, 2009). Although a subset of this diversity can be linked directly to defined phylogenetic groups with cultivated representatives, an increasing number of unaffiliated groups known as candidate phyla or divisions have emerged. Currently, 45 candidate phyla are recognized in public databases on the basis of 16S rRNA gene sequence information, although it is likely that there are many more phyla that have yet to be formally recognized (Rappé and Giovannoni, 2003; McDonald et al., 2011). One of the most prevalent candidate phyla identified in studies of marine microbial diversity is the bacterial Marine Group A (MGA; Fuhrman et al., 1993; Gordon and Giovannoni, 1996; Fuchs et al., 2005; DeLong et al., 2006; Stevens and Ulloa, 2008; Schattenhofer et al., 2009). The first representatives of MGA were described and named as ‘Marine Group A’ (Fuhrman et al., 1993; Fuhrman and Davis, 1997) or the ‘SAR406 gene lineage’ (Gordon and Giovannoni, 1996) based on 16S rRNA gene sequence information collected from Atlantic and Pacific Ocean waters. Contemporary phylogenetic analyses indicate that MGA is most closely related to the phylum Caldithrix, named after a genus of anaerobic, mixotrophic, thermophiles obtained from a hydrothermal vent chimney in the Mid-Atlantic Ridge (Miroshnichenko et al., 2003; Rappé and Giovannoni, 2003).

MGA are most prevalent below the photic zone in stratified waters with distinct halo or oxyclines. Indeed, in surveys of oxygen minimum zones (OMZs) and permanent or seasonally stratified anoxic basins, 16S rRNA gene sequences affiliated with MGA are well represented in clone libraries (Madrid et al., 2001; Fuchs et al., 2005; Stevens and Ulloa, 2008; Zaikova et al., 2010; Wright et al., 2012). In these systems, O2 serves as a key organizing principle for microbial community structure and function, defining specific metabolic niches and biogeochemical potentials across the oxycline (Wright et al., 2012). MGA are particularly diverse and abundant within the OMZ of the Northeast subarctic Pacific Ocean (NESAP; Wright et al., 2012). The distinct and well-studied coastal to open-ocean gradients of biological production, nutrients and O2 existing within the OMZ of the NESAP make it an ideal natural laboratory in which to explore ecological and biogeochemical roles of MGA in the ocean. Here, we use a combination of catalyzed reporter deposition fluorescence in situ hybridization (CARD-FISH), 16S rRNA gene clone libraries and pyrotag sequencing to quantify MGA abundance and diversity along the Line P oceanographic transect of the NESAP (Pena and Bograd, 2007). We then apply statistical analyses to explore the hypothesis of O2 and other environmental factors as drivers of habitat selection for different MGA subgroups in the water column.

Materials and methods

Sample collection and processing

Sampling was conducted via multiple hydrocasts using a conductivity, temperature, depth rosette water sampler aboard the CCGS John P Tully during Line P cruise 2009-09 in the NESAP in June 2009 (major stations: P4 (48°39.0°N, 126°4.0°W)—7 June, P12 (48°58.2°N, 130°40.0°W)—9 June and P26 (50°N, 145°W)—14 June). At these three stations, large volume (20 l) samples for DNA isolation were collected from the surface (10 m), whereas 120 l samples were taken from three depths spanning the OMZ core and upper and deep oxyclines (500 m, 1000 m and 1300 m at station P4 and 500 m, 1000 m and 2000, m at stations P12 and P26). Sample collection and filtration protocols can be viewed as visualized experiments at http://www.jove.com/video/1159/ (Zaikova et al., 2009) and http://www.jove.com/video/1161/ (Walsh et al., 2009a), respectively.

For small-volume sampling (for CARD-FISH), the water from Niskin bottles was transferred into pre-rinsed 1-l plastic bottles, filtered through a 10-μm nylon mesh filter and processed immediately. The conductivity, temperature, depth-mounted O2 probe (Model SBE 43, Sea-Bird Electronics, Bellevue, WA USA) reported O2 concentrations in μmol kg−1. Nutrient samples were collected in plastic tubes and analyzed at sea (stored at 4 °C and in the dark before analysis) using an Astoria Analyzer (Astoria-Pacific, Clackamas, OR, USA) as described by Barwell-Clarke and Whitney (1996).

Chlorophyll a

Chlorophyll a (Chla) was measured in situ with a Seapoint chlorophyll fluorometer (Seapoint Sensors, Exeter, NH, USA) and ground-truthed with 109 selected reference samples collected on 47 mm GF/F filters (Whatman International, Maidstone, UK) for Chla extraction (Holm-Hansen et al., 1965). The linear regression between reference sample fluorescence and Chla data were used to transform depth corrected fluorescence units to Chla (Cuttelod and Herve, 2010; R2=0.90; data not shown).

Enumeration of cells by flow cytometry

Cells were enumerated by flow cytometry using samples fixed with formaldehyde (final concentration of 4% wt/vol) and stored at 4 °C until analysis using SYBR Green I (Invitrogen, Carlsbad, CA, USA) on a FACS LSRII (Becton Dickonson, Franklin Lakes, NJ, USA; Zaikova et al. 2010).

Catalyzed reporter deposition fluorescence in situ hybridization

Pre-filtered (10 μm) seawater samples were fixed with formaldehyde (16%, Polysciences, Warrington, PA, USA) at a final concentration of 1–2% at 4 °C for 12–24 h. Subsamples were filtered onto 47 mm 0.2 μm membrane filters (GTTP, Millipore, Billerica, MA, USA) and rinsed with Milli-Q water. Filters were left to air dry and then stored at −80 °C until analysis by CARD-FISH as described by Pernthaler et al. (2004). In brief, cells were fixed to the filter membrane by agarose embedding. Endogenous peroxidases were inactivated by HCl treatment, cells were permeabilized by lysozyme (for probes EUBI-III (Amann et al., 1990; Daims et al., 1999), NON338 (Wallner et al., 1993), SAR406-97 (Fuchs et al., 2005)) or a combination of lysozyme and achromopeptidase or HCl (tested for optimization only). For hybridization, horseradish peroxidase-labeled probes EUBI-III and NON338, and SAR406-97 were added to hybridization buffers containing 35% and 40% formamide (Fisher, Pittsburg, PA, USA), respectively. Hybridizations were performed at 46 °C and followed by washing steps to remove unspecifically bound probe. During the CARD step, the dye Alexa Fluor488 (Invitrogen, Molecular Probes, Carlsbad, CA) was combined with the remaining substrate mix at 1:300. The fraction of FISH-stained bacteria was quantified microscopically at × 1000 magnification in at least 1000 4′6-diamidino-2-phenyl indole-stained cells in 10 or more fields of vision per sample using an AxioImager (Zeiss, Jena, Germany).

Environmental DNA extraction for 16S rRNA gene clone library construction

DNA was extracted from sterivex filters as described in Zaikova et al. (2010) and DeLong et al. (2006). The DNA extraction protocol can be viewed as a visualized experiment at http://www.jove.com/video/1352/ (Wright et al., 2009).

PCR amplification of 16S rRNA gene, clone library construction and sequencing

A total of 12 DNA extracts from large volume samples collected from four depths at stations P4, P12 and P26 in February 2009 (using the same sampling plan and protocols described above) were amplified using small subunit ribosomal DNA (16S rRNA gene) primers targeting the bacterial domain: B27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and U1492R (5′-GGTTACCTTATGTACGACTT-3′) under the following PCR conditions: 3 min at 94 °C followed by 35 cycles of 94 °C for 40 s, 55 °C for 1.5 min, 72 °C for 2 min and a final extension of 10 min at 72 °C. Each 50 μl reaction contained 1 μl of DNA, 1 μl each 10 mM forward and reverse primer, 2.5 U Taq (Qiagen, Germantown, MD, USA), 5 μl 10 mM deoxynucleotides and 41.5 μl 1X Qiagen PCR Buffer. 16S rRNA gene amplicons were purified, transformed and cloned as described previously (Zaikova et al., 2010-3′) with the following modifications: one 384-well plate per depth interval was picked and sent for Sanger sequencing at the Michael Smith Genome Sciences Centre (Vancouver, British Columbia, Canada). Sequence data were collected on an AB 3730xls (Applied Biosystems, Carslbad, CA, USA). Plasmids were sequenced bidirectionally with M13F (5′-GTAAAACGACGGCCAG-3′) and M13R (5′-CAGGAAACAGCTATGAC-3′) primers. Bidirectional sequence reads were assembled using Sequencher v4.8 (Gene Codes Corporation, Ann Arbor, MI, USA) and manually edited for base-calling errors. The resulting data sets were checked for chimeras with the open source application Bellerophon (Huber et al., 2004; using default settings) and 745 chimeric sequences were removed.

Phylogenetic analysis and tree construction using MGA 16S rRNA gene sequences

A total of 3164 non-chimeric 16S rRNA gene sequences were imported into the ARB software package (release 106; Ludwig et al., 2004). Sequences were added to the full-length SILVA database (http://www.arb-silva.de; Pruesse et al., 2007), aligned to the closest relative and added to an existing tree of sequences from the ARB database by using the ARB parsimony tool (using default parameters).

A maximum likelihood phylogenetic tree of MGA 16S rRNA gene sequences exported from ARB was inferred by PHYML (Guindon et al., 2005) using an HKY+4G+I model of nucleotide evolution where the parameter of the gamma distribution, the proportion of invariable sites and the transition/transversion ratio were estimated for each data set. The confidence of each node was determined by assembling a consensus tree of 100 bootstrap replicates. Bacterial 16S rRNA gene sequences (including 170 previously published sequences) generated from the Line P transect in June 2008 (station P4 1000 m; Walsh et al., 2009b) were also placed in taxonomic hierarchy for downstream analysis using the NAST aligner (DeSantis et al., 2006b) and blast using default parameters against the 2008 Greengenes database (DeSantis et al., 2006a), and 290 sequences were identified as belonging to MGA. These 290 sequences were clustered at 97% identity using mothur (v.1.19.0; Schloss et al., 2009). Representative sequences from each of these clusters were identified using the get.oturep command in mothur and were included in the phylogenetic tree.

PCR amplification of 16S rRNA gene for pyrotag sequencing

To more directly compare the quantitative distribution of MGA in relation to CARD-FISH counts, the V6–V8 region of 16S rRNA was amplified from June 2009 DNA samples using primers 926F (5′-cctatcccctgtgtgccttggcagtctcag AAACTYAAAKGAATTGRCGG-3′) and 1392R (5′-ccatctcatccctgcgtgtctccgactcag-<XXXXX>-ACGGGCGGTGTGTRC-3′). Primer sequences were modified by the addition of 454A or B adapter sequences (lower case). In addition, the reverse primer included a 5-bp barcode designated <XXXXX> for multiplexing of samples during sequencing. Twenty-microlitre PCR reactions were performed in duplicate and pooled to minimize PCR bias using 0.4 μl Advantage GC 2 Polymerase Mix (Advantage-2 GC PCR Kit, Clonetech, Mountainview, CA, USA), 4 μl 5X GC PCR buffer, 2 μl 5 M GC Melt Solution, 0.4 μl 10 mM dNTP mix (MBI Fermentas, Glen Burnie, MA, USA), 1.0 μl of each 25 nM primer and 10 ng sample DNA. The thermal cycler protocol was 95 °C for 3 min, 25 cycles of 95 °C for 30 s, 50 °C for 45 s, and 68 °C for 90 s and a final 10-min extension at 68 °C. PCR amplicons were purified using SPRI Beads and quantified using a Qubit fluorometer (Invitrogen). Samples were diluted to 10 ng μl−1 and mixed in equal concentrations. Emulsion PCR and sequencing of the PCR amplicons were performed at the Department of Energy Joint Genome Institute (Walnut Creek, CA, USA) following the Roche 454 GS FLX Titanium (454 Life Sciences, Branford, CT, USA) technology according to the manufacturer’s instructions.

Processing of pyrotag sequences

A total of 219 610 pyrotag sequences were analyzed using the Quantitative Insights Into Microbial Ecology (QIIME) software package (Caporaso et al., 2010). Reads with length <200 bases, ambiguous bases and homopolymer runs were removed before chimera detection. Chimeras were detected using the chimera slayer provided in the QIIME software package and removed before taxonomic analysis. A total of 212 611 non-chimeric sequences were phylogenetically identified in QIIME using a BLAST-based assignment method and clustered at 97% identity against the Greengenes taxonomic database (DeSantis et al., 2006a). Singleton operational taxonomic units (OTUs; represented by one read) were omitted from downstream analyses, as recommended by Kunin et al. (2010), Tedersoo et al. (2010) and Gihring et al. (2012), leaving 183 212 sequences for downstream analysis.

Clustering of pyrotags to 16S rRNA gene clone library sequence clusters

To resolve patterns of distribution among MGA clusters as a function of geographic location in the NESAP, pyrotag sequences were recruited to MGA 16S rRNA gene clone library sequence clusters using a 97% identity cutoff in mothur. Blastn was used to query 183 212 pyrotags against a database containing 290 16S rRNA gene clone library sequences assigned to MGA based on Greengenes taxonomy. Only hits with a perfect match across the full length of a query sequence were retrieved, and the number of pyrotags mapping to all sequences in each cluster was summed. If a pyrotag mapped to >1 cluster, its relative contribution to each cluster was calculated by dividing by the number of clusters it mapped to and assigning the relevant fraction to each cluster. The number of pyrotags mapping to each cluster was normalized to the total number of bacterial tags in each sample (Table 1) and visualized as a bubble plot using bubble.pl, available for download at http://www.cmde.science.ubc.ca/hallam/bubble.php. A rarefaction curve for full-length MGA 16S rRNA sequences and MGA pyrotag sequences was calculated and plotted using QIIME (Caporaso et al., 2010).

Table 1 Chemical and biological parameters at Line P stations P4, P12 and P26 in June 2009
Full size table

Estimating probe SAR406-97 detection efficiency

To test the predicted maximum binding efficiency of probe SAR406-97 (Fuchs et al., 2005; 5′-CACCCGTTCGCCAGTTTA) against MGA 16S rRNA gene clone library sequences from the NESAP, blastn (E-value=1000, word_size=7) was used to query the probe sequence against the 290 16S rRNA gene clone library sequences assigned to MGA based on Greengenes taxonomy and collect all local alignments with similarity to the probe sequence. Probe efficiency was described using the percentage of MGA sequences that contained local alignments to the probe across a range of E-value scores for each cluster.

Results

Physicochemical characteristics of the NESAP

Our study site, Line P, is a 1425-km survey line of the NESAP, originating in Saanich Inlet, British Columbia, Canada (SI; 48°N, 123°W), and terminating at Ocean Station Papa (also known as station P26; 50°N, 145°W), on the southeast edge of the Alaskan Gyre (Pena and Bograd, 2007; Pena and Varela, 2007; Supplementary Figure S1). The NESAP is characterized by strong stratification with a maximum winter mixing depth of 125–150 m (Freeland, 1997; Whitney et al., 1998). As such, the interior regions of the NESAP are insulated from the atmosphere, creating a vast OMZ centered at 1000 m with oxyclines extending from ∼400 m to 2000, m containing O2 concentrations ranging between ∼9 μmol kg−1 and 60 μmol kg−1 (Freeland, 1997; Whitney et al., 2007). The NESAP OMZ (also referred to as the Eastern Subtropical North Pacific OMZ) is the largest and least studied permanent OMZ in the global ocean (Paulmier and Ruiz-Pino, 2009).

Relevant physicochemical data from representative coastal (P4), transition (P12) and open-ocean (P26) stations measured along the Line P transect and related to this study are described below. Salinity gradients ranging from 32.2 PSU to 32.6 PSU at the surface (10 m) and 34.1–34.6 PSU in the ocean’s interior generated a stratified water column across the Line P transect (Supplementary Figure S2). Chla was present in the top ∼100 m, with deep chlorophyll maxima ranging from 0.5 mg l−1 at 41 m depth at P26 to 1.1 mg l−1 at 25 m depth at P4 (Figure 1). Average O2 concentrations were 302 μmol kg−1 at the surface, reaching a minimum of 8.6–15 μmol kg−1 between 1000 m and 1100 m across the transect (Table 1, Figure 2). The OMZ core (defined as O2<20 μM (∼19.5 μmol kg−1); Helly and Levin, 2004; Paulmier and Ruiz-Pino, 2009) was 766±73 m thick and centered at 1026±63 m. Nutrient concentrations were higher in the OMZ core and the upper (500 m) and deep (2000, m) oxyclines than at the surface (Table 1, Supplementary Figure S2). In 10 m samples, nitrate and phosphate concentrations were highest at P26 (9.9 μmol l−1 and 1.0 μmol l−1, respectively). At 1000 m, nitrate concentration was highest at P26 (47.5 μmol l−1), whereas phosphate concentration was highest at P4 (3.3 μmol l−1). All contextual data are available through the Canadian Department of Fisheries and Oceans (http://www.pac.dfo-mpo.gc.ca/science/oceans/data-donnees/line-p/).

Figure 1
figure 1

Contextual data for Line P stations P4, P12, and P26 in June 2009. Depicted are Chla, temperature, and total cell counts detected by flow cytometry. A full-colour version of this figure is available at The ISME Journal Online.

Full size image
Figure 2
figure 2

Relative abundance of MGA by CARD-FISH in the NESAP at Line P stations P4, P12 and P26 in June 2009. O2 concentration is depicted as colored background and MGA abundance is overlaid as gray bubbles.

Full size image

Microbial cell numbers

Total microbial abundance along the Line P transect was (1.3±0.1) × 105 ml−1 in surface waters and (1.39±0.2) × 104 ml−1 in waters >200 m as measured by flow cytometry (Table 1, Figure 1). The overall detection of Bacteria by probes EUBI-III ranged from 25.5%±7.6% to 79.5%±8.6% of total 4′6-diamidino-2-phenyl indole cell counts with higher detection rates in surface samples (Supplementary Table S1). Low EUB detection did not appear to result from poor cell lysis, as comparison of lysozyme vs lysozyme/achromopeptidase treatment (Pernthaler et al., 2004) revealed no significant differences (data not shown). Sequence comparison by BLAST analysis suggested that >90% of our full-length bacterial 16S rRNA gene clone library sequences were targeted by EUBI-III probes with an E-value of 10–4 (corresponding to a blastn result with no mismatches and up to one missing 3′ base; data not shown).

Diversity and population structure of MGA

Relative abundance of MGA cells as detected by probe SAR406-97 was similar at stations P4, P12 and P26, with minima in surface waters and maxima in waters ≥500 m (⩽1.3% vs ∼8%, respectively; Figure 2). At stations P12 and P26, MGA abundance peaked in the core of the OMZ (6.7%±1.8% and 8.2%±1.6%, respectively) with lower values (3.3–5.5%) in the upper and deep oxyclines (Table 1, Figure 2). At station P4, MGA abundance peaked in the upper oxycline (7.8%±2.3%) and decreased throughout the OMZ core and deep oxycline. Blastn-based sequence comparisons against our full-length 16S rRNA gene clone library sequences suggested that probe SAR406-97 targeted ∼76% of all MGA sequences (see below) with an E-value of 10–4 (corresponding to a blastn result with no mismatches and up to one missing 3′ base; Supplementary Tables S2a and b).

A total of 290 MGA 16S rRNA gene sequences were recovered from 3164 bacterial sequences traversing the water column at stations P4, P12 and P26. MGA sequences comprised an average of 0.7%±0.84% of 10 m clone libraries and 11.2%±3.9% of libraries from O2-deficient waters (<90 μmol kg−1 O2) with a maximum of 16.4% at P26 1000 m (Table 1). MGA 16S rRNA gene sequences clustered at 97% identity into 121 distinct OTUs, 97 of which contained only singletons (Supplementary Table S2). Representative sequences obtained for each OTU were placed in phylogenetic context with relevant reference sequences (Figure 3). Five previously defined subgroups were recovered (ZA3648c and ZA3312c (Fuchs, unpublished), Arctic96B-7 and Arctic95A-2 (Bano and Hollibaugh, 2002) and SAR406 (Gordon and Giovannoni, 1996)), and five additional subgroups were defined (HF770D10, P262000D03, P41300E03, P262000N21 and A714018). The most abundant OTUs present along the Line P transect comprised between 1% and 4% of at least one clone library and belonged to subgroups Arctic95A-2, HF770D10, SAR406, Arctic96B-7 and ZA3312c (Figure 4, Supplementary Table S2a).

Figure 3
figure 3

Unrooted phylogenetic tree based on 16S rRNA gene clone sequences showing the phylogenetic affiliation of MGA sequences identified in this study. The tree was inferred using maximum likelihood implemented in PhyML (Guindon et al., 2005). Reference sequences from other environments are marked with an asterisk. The bar represents 10% estimated sequence divergence.

Full size image
Figure 4
figure 4

Relative abundance of MGA pyrotags affiliated with full-length MGA 16S rRNA gene clone OTUs recovered from the Northeast subarctic Pacific Ocean. Black circles represent proportion of bacterial pyrotags affiliated with each 16S rRNA OTU in each sample. A full-colour version of this figure is available at The ISME Journal Online.

Full size image

To explore the diversity and population structure of MGA subgroups with increased resolution, we performed 454-pyrotag sequencing (Table 1). Pyrotags affiliated with MGA OTUs were identified using two approaches: (1) recruitment of pyrotags to full-length 16S rRNA gene sequences and (2) direct taxonomic assignment of pyrotags in blast-based queries to identify OTUs not detected in clone libraries.

In the first approach, we recruited all pyrotags to all 16S rRNA gene clone library sequences affiliated with MGA (see Materials and methods). A total of 4403 pyrotags formed identical matches to 78 out of 121 previously defined MGA OTUs (Figure 4). The relative proportion of bacterial pyrotags affiliated with MGA OTUs ranged from ∼0.01% in 10 m samples to a maximum of 5.7% at P4 1000 m. Within O2-deficient waters, the average proportion of bacterial pyrotags belonging to MGA was 4.4%±0.73%. The most abundant MGA OTUs based on pyrotag recruitment were affiliated with Arctic95A-2 (∼2.4%), Arctic96B-7 (0.55%), SAR406 (∼0.45%), HF770D10 (0.55%) and A714018 (0.26%).

In the second approach, all non-singleton pyrotags were queried against the Greengenes database (DeSantis et al., 2006a) resulting in the identification of 10 278 sequences affiliated with MGA (Figure 5a). The relative proportion of bacterial pyrotags affiliated with MGA ranged from ∼0.1% in 10 m samples to a maximum of 11.6% at P4 1000 m (Table 1). Within O2-deficient waters, the average proportion of bacterial pyrotags belonging to MGA was 9.9%±1.8%. To identify MGA OTUs unique to pyrotags, we extracted the corresponding V6–V8 region from the 290 16S rRNA gene clone library sequences identified as MGA and clustered these with the subset of pyrotag sequences affiliated with MGA at 97% identity into 566 distinct OTUs, 491 of which were unique to pyrotags (Figure 5b). However, the majority of abundant OTUs (containing >200 sequences) were common between 16S rRNA gene clone libraries and pyrotag data sets (Figure 5c). Of the unique pyrotag OTUs, 249 were non-singleton and contained 4253 pyrotags (40% of MGA pyrotags), with the most abundant OTU containing 1409 sequences (13.3% of MGA pyrotags; Figure 5c). The slope of the rarefaction curve for MGA pyrotags became nearly asymptotic, indicating that the ultimate richness of MGA OTUs was very nearly sampled (Supplementary Figure S3). In contrast, the rarefaction curve for MGA 16S rRNA gene clone library sequences indicated incomplete sampling.

Figure 5
figure 5

Comparison of V6–V8 region of full-length 16S rRNA gene clone sequences affiliated with MGA, and pyrotags taxonomically identified as MGA by comparison with Greengenes. (a) Number of MGA sequences shared between and unique to 16S rRNA gene clone libraries and pyrotags. (b) Number of MGA OTUs shared between and unique to 16S rRNA gene clone libraries and pyrotags. (c) Sequence distribution within shared and unique MGA OTUs.

Full size image

Comparing MGA abundance across methods

To evaluate consistency in estimating MGA abundance using CARD-FISH, 16S rRNA gene clone libraries and pyrotags, Spearman’s rank correlation coefficients (ρ) were determined (Supplementary Figure S4). CARD-FISH abundance estimates were significantly correlated (P<0.05) with 16S rRNA gene clone library sequence abundance (ρ=0.755) but not with pyrotag sequence abundance (ρ=0.469; Supplementary Figures S4a and b). 16S rRNA gene clone library and pyrotag sequence abundance were also significantly correlated (ρ=0.580, Supplementary Figure S4c).

To explore potential drivers of MGA habitat selection, we calculated Spearman’s rank correlation coefficients between CARD-FISH, 16S rRNA gene clone library, and pyrotag sequence abundance and environmental parameters. When calculated across the entire transect, the abundance of MGA as estimated by CARD-FISH was significantly correlated with decreasing temperature, O2 and Chla, and increasing nitrate, phosphate and silicate (Table 2). However, when correlations were calculated for each station independently, statistically significant correlations were only identified at station P26 where MGA abundance was more strongly correlated with decreasing O2 and increasing nitrate and phosphate concentrations than with temperature, Chla or silicate (Table 2, Supplementary Figure S5). When calculated across the entire transect and each station independently, the relative abundance of MGA OTUs based on 16S rRNA gene clone library sequences was not significantly correlated with environmental parameters (data not shown). However, the relative abundance of four OTUs identified in pyrotags showed significant correlations across the entire transect with decreasing O2 after a Bonferroni correction was applied (P<0.000079; Table 3). OTUs significantly correlated with decreasing O2 were affiliated with two subgroups of MGA (Arctic95A-2 and A714018), and an additional 13 OTUs affiliated with HF770D10, ZA3648c, Arctic96B-7, Arctic95A-2, SAR406 and A714018 were weakly correlated (P<0.05; Table 3). In addition, out of all 78 MGA OTUs identified by binning pyrotags to full-length 16S rRNA gene sequences, 10 displayed significant correlations (P<0.000079) with increasing depth, salinity and nutrients (nitrate, phosphate, silicate) or decreasing Chla (Supplementary Table S3).

Table 2 Spearman’s rank correlation coefficients between relative abundance of MGA estimated by CARD-FISHa and environmental parameters
Full size table
Table 3 Pyrotag OTUs with statistically significant Spearman’s rank correlations (ρ) with oxygen concentration (17 out of 79) in the NESAP
Full size table

Discussion

MGA abundance estimates in the NESAP were highly correlated between CARD-FISH and 16S rRNA gene clone library, but not between CARD-FISH and pyrotag sequences, whereas 16S rRNA gene clone library and pyrotag sequences were correlated based on Spearman’s rank correlations. Moreover, CARD-FISH-based estimates were consistently <16S rRNA gene clone library or pyrotag sequence estimates for the same samples. For example, the average relative abundance of MGA sequences in O2-deficient waters was 11.0%±3.9% based on 16S rRNA gene clone libraries, 9.9%±1.8% based on pyrotags and 5.6%±1.9% based on CARD-FISH (Table 1). This suggests an under or overestimation of MGA abundance by one, some or all of the methods used. The discrepancy between methods could be purely based on primer and probe differences and the underlying methods applied. One perspective would be that CARD-FISH with probe SAR406–97 underestimated MGA abundance. Lower detection efficiency by CARD-FISH has been attributed to limited probe access to target cells when using horseradish peroxidase-labeled probes (Schoenhuber et al., 1997), even after careful permeabilization optimization (Woebken et al., 2007). Also, the permeabilization step might cause leakage of ribosomes from target cells, which in turn could result in low-ribosome content cells dropping below the CARD-FISH detection limit (Hoshino et al., 2008). Alternatively, MGA subgroups could harbor variable copy numbers of the 16S rRNA gene, inflating PCR-based metrics (Acinas et al., 2004).

Rarefaction curves for MGA 16S rRNA gene clone library and pyrotag sequences recovered from the NESAP were consistent with known methodological limitations based on variable sample size and potential primer bias (Engelbrektson et al., 2010; Schloss et al., 2011; Gihring et al., 2012). Clustering the combined data sets enabled pyrotag assignments to 78 out of 121 OTUs defined by 16S rRNA gene clone library sequences. The inability to assign pyrotags to all 121 OTUs may have resulted from the conservative nature of our clustering method: we required full-length pyrotag sequences to match a cognate 16S rRNA gene clone library sequence with no mismatches. Alternatively, it is possible that time variable patterns in the abundance of MGA OTUs prevented assignment of all June 2009 pyrotags to OTUs identified in February 2009 16S rRNA gene clone libraries. Although ∼50–75% of pyrotags identified as MGA in blast-based taxonomic queries were not assigned to OTUs defined by 16S rRNA gene clone library sequences, pyrotags affiliated with all 10 MGA subgroups were recovered. Indeed, comparison of 16S rRNA gene clone library and pyrotag sequence clusters revealed that the majority of MGA sequences (57%) and abundant MGA OTUs (containing >200 sequences) were identified using both methods (Figure 5). Unique pyrotag OTUs were generally composed of <50 sequences with a single abundant OTU containing 1490 pyrotags that could not be assigned to defined MGA subgroups. Sequences in this OTU were recovered from 500 m, 1000 m, 1300 m and 2000, m samples at all three stations indicating an environmental origin. The extent to which unique pyrotag OTUs captured components of the ‘rare biosphere’ (Sogin et al., 2006) subject to time-variable changes in population structure remains to be determined. Despite this uncertainty, the recovery of a single abundant OTU unaffiliated with MGA subgroups defined by 16S rRNA gene clone library sequences suggests that the majority of abundant MGA subgroups in the NESAP have been identified.

Spearman’s rank correlation coefficients provided statistical support for vertical partitioning of MGA subgroups in the NESAP water column. The relative abundance of MGA OTUs identified in pyrotags (affiliated with Arctic95A-2 and Arctic96B-7) displayed a negative correlation with O2 concentration consistent with habitat selection within suboxic waters (1–20 μmol kg−1) of the OMZ. The extent to which patterns of vertical partitioning among and between MGA OTUs represent ecological types (ecotypes; Koeppel et al., 2008) or class divisions remains to be determined. Environmental gradients are common drivers of selection among microorganisms at different ecological scales. For example, Johnson et al. (2006) documented niche partitioning of Prochlorococcus ecotypes over ocean-basin scales across temperature (eMED4 vs eMIT9312) and nutrient (eNATL2A or eMIT9313) gradients. Similarly, SAR11 ecotypes display depth-specific distributions with subclade Ia members more prevalent in the euphotic zone and subclade II members more abundant in deeper (mesopelagic) waters (Field et al., 1997). Such distribution patterns are associated with changes in genome composition that promote differential fitness including allelic variation (Urbach and Chisholm, 1998; Urbach et al., 1998; Wilhelm et al., 2007; Zhao and Qin, 2007) and metabolic island formation (Rocap et al., 2003; Coleman et al., 2006; Coleman and Chisholm, 2007; Kettler et al., 2007;Wilhelm et al., 2007).

Looking forward, genome-scale sequence data (that is, single-cell and metagenomic data) representative of defined MGA subgroups will be invaluable both to more accurately assess evolutionary relationships between MGA and thermophilic bacteria, such as Caldithrix, as well as to attach metabolic repertoires to defined MGA subgroups (Shapiro et al., 2012; Swan et al., 2011). In turn, metabolic characterization of MGA subgroups will assist in determining whether observed 16S rRNA-based patterns of distribution across the oxycline are associated with variable forms of energy metabolism, consistent with redox-driven niche partitioning and ecotype differentiation. In addition, more extensive quantitative studies documenting the temporal dynamics of extant MGA subgroups across multiple provinces are needed to assess the stability of MGA population structure and function and better constrain the ecological and biogeochemical roles of MGA within OMZs.

Accession numbers

Bacterial 16S rRNA sequences reported in this study were deposited in GenBank with the accession numbers HQ242143–HQ242376 and HQ671746–HQ674628. Bacterial 16S rRNA sequences previously published in Walsh et al. (2009b) can be found under the accession numbers GQ351133–GQ351265. Pyrotag sequences reported in this study were deposited in GenBank with the accession number SRA051605.