Introduction

In the ecology of macroorganisms, ecosystem multifunctionality has been related to species diversity, with some degree of redundancy ‘built in’ to community structure (Hector and Bagchi, 2007). However, the high abundance of microorganisms has previously precluded sampling at a scale considered robust enough to reliably estimate community richness. Hence, ecological theories pertaining to the impact of microbial diversity on ecosystem function are currently poorly informed. The recent development of ribosomal tag pyrosequencing methods has allowed the collection of more detailed organismal abundance data for the analysis of community composition and diversity. These methods are capable of detecting both the most abundant community members and those less-abundant components that are seldom, if ever, retrieved by traditional cultivation or cloning and sequencing methods (Pedrós-Alió, 2006). This is an important development in microbial ecology, because, although abundant and easily detectable organisms may have an obvious impact on energy harvesting and nutrient cycling, the role of less-abundant organisms is not easily understood. These rare organisms may have the following important environmental roles: they may act as keystone species in complex microbially mediated cycles, serve as a reservoir of genetic, and functional diversity and/or buffer ecosystems against species loss or environmental change. As technology has improved, the roles of these rare organisms have been further appreciated; for example, Montoya et al. (2004) showed the importance of numerically unimportant unicellular diazatrophs in nitrogen fixation in the oligotrophic ocean, whereas Musat et al. (2008) showed that, despite representing only 0.3% of the total community, Chromatium okenii was potentially responsible for 40% and 70% of the total ammonium and carbon uptake, respectively, in the lake system sampled.

Adequate baseline data are required to set the framework against which to gauge community shifts in response to global climate change and other anthropogenic perturbation. Although Eucarya–Bacteria–Archaea interactions are common in the ocean and undoubtedly have a major role in determining community structure (Fuhrman and Hagstrom, 2008), these groups have rarely been analyzed together in environmental surveys. There is a growing recognition of microbial eukaryotes as primary producers, bacterial grazers and parasites in the ocean (Massana and Pedrós-Alió, 2008). Determining how the community structure of each domain changes over spatial and temporal scales is key to developing a more holistic overview of marine ecosystem dynamics. Recent pyrosequencing surveys of Bacteria and Archaea in deep-sea pelagic and benthic communities have highlighted an ‘unexpected’ diversity in these biomes (Sogin et al., 2006; Huber et al., 2007), whereas metagenomic analyses have revealed a remarkable functional diversity (Venter et al., 2004; DeLong et al., 2006). A recent meta-analysis of marine pyrosequencing and metagenomic data concluded that planktonic bacterial communities in the upper ocean were less diverse than those around hydrothermal vents in the deep sea and soil communities (Quince et al., 2008). However, ribosomal tag sequencing analysis has not been performed over a depth profile at any location, and no pyrosequencing or metagenomic survey to our knowledge has focused on all three domains in the marine environment. To address this paucity of data, we examined the community structure of Bacteria, Archaea and micro-Eucarya along a depth gradient at Station ALOHA, an open-ocean oligotrophic site in the sub-tropical North Pacific. The depths sampled represent the epi-, meso- and bathypelagia, each of which has distinct physical and biochemical characteristics, and potentially different scales of spatial and temporal variability (Hewson et al., 2006a, 2006b). Differences in the Bacteria and Archaea functional gene content have been observed across this depth gradient (DeLong et al., 2006); however, in that study, samples from different depths were collected at different times and may thus be temporally confounded (Hewson et al., 2006b). How phylogenetic structure changes throughout the water column and what factors influence community diversity are still debated. Here, we deconstruct the composition of each domain at different levels of phylogenetic resolution, analyzing how taxonomy and community structure change over this depth profile. We also compare phylogenetic data with those from previous marine bacterioplankton surveys, and compare the diversity of marine Bacteria with that found in soils using similar survey methods.

Materials and methods

Water body parameters

Information on water body characteristics, such as counts of microbial cells, was taken from the HOT-DOGS website (http://hahana.soest.hawaii.edu/hot/hot-dogs/interface.html) of the Hawaii Ocean Time-Series (HOT).

Sampling, assay preparation and sequencing

Seawater (4 l) from depths of 10, 800 and 4400 m was collected from the HOT Station ALOHA (22.45 °N, 158 °W) during HOT cruise 169 (May 16–20, 2005) and was sequentially filtered through GFA glass fiber filters (Whatman, Clifton, NJ, USA) before passage through 0.22 μm HV Durapore (Millipore, Bedford, MA, USA) filters. Sample preparation and DNA extraction protocols have been reported previously (Brown and Donachie, 2007) and are summarized in the Supplementary information. PCR primer sequences used to amplify the ribosomal sequence tags (Table 1) used slightly modified common 16S/18S rDNA specific primers 1392F and 1492R (Lane, 1991), amended with 454 Life Sciences A and B adapters. A full discussion of primer specificity is supplied in the Supplementary materials. We calculated that primers would match to >80% of known phylogenetic diversity across all three domains. DNA extracted from both the Whatman and Durapore filters for the same depths was pooled, and ∼5 ng was used as template in PCRs. Each PCR reaction (100 μl volumes) contained 5U AmpliTaq Gold (Applied Biosystems, Carlsbad, CA, USA), 1 × PCR Amplitaq Gold buffer (Applied Biosystems), 2.5 mM MgCl2, 400 μM of each deoxynucleotide and 400 nM of each primer. Reaction mixtures were held at 94 °C for 10 min, followed by 24 amplification cycles of 94 °C for 40 s, 55 °C for 40 s and 72 °C for 40 s, with a final step at 72 °C for 2 min. Amplification products from three separate reactions for each sample were pooled and purified in a MinElute PCR purification kit (Qiagen, Valencia, CA, USA).

Table 1 Primer sequences used in this study
Full size table

Pyrosequencing data that were returned from 454 Life Sciences were subject to a systematic check to remove low-quality reads. We eliminated (i) sequences that did not perfectly match the 3 bp key code and primer sequence at the beginning of the read, (ii) sequences that did not perfectly match at least the first 10 bp of the distal primer, (iii) sequences that contained any undetermined nucleotides (N) and (iv) sequence reads of <50 bp after the removal of both primers. Use of these quality control measures greatly increases the resultant read accuracy in pyrosequencing assays (Huse et al., 2007). Quality sequences were compared by BLAST (Altschul et al., 1990) (e-value cutoff of 10−5) with the GenBank nr database, and the domain of origin was determined from the best high-scoring pair. A further extensive phylogenetic classification was carried out as described in the Supplementary materials. BLAST phylotypes were analyzed to identify those statistically more likely to be found at one depth (overrepresented), following the resampling method of Rodriguez-Brito et al. (2006).

Community comparison

We clustered and compared samples using the nonalignment clustering algorithms in the CD-HIT package (downloaded from http://www.bioinformatics.org/cd-hit/) (Li and Godzik, 2006). Archaea, Bacteria and Eucarya databases, containing all sequences from each depth, were compared across all similarity thresholds between 0.8 and 1 using the program CD-HIT-EST-2D (downloaded from http://www.bioinformatics.org/cd-hit/), with word sizes set according to the manual (http://cd-hit.org/). This is a quantitative two-way comparison (for example, the percentage of all sequences from 800 m that cluster with sequences from 10 m at a given percentage sequence similarity is determined, followed by the reverse), and thus this method takes into account sequence abundance as well as identity. A detailed description of the derivation of parametric and nonparametric diversity estimators, based on frequency count data from CD-HIT clustering, and choice of models used is available in the Supplementary materials

Results

Hydrographic considerations

Compartments of the water column investigated correspond to the epi- (10 m), meso- (800 m) and bathypelagia (4400 m) in the Sub-Tropical North Pacific Gyre, and differ dramatically in their physical and chemical characteristics (Supplementary Figure S1). The availability of photosynthetically active radiation drives the growth of photosynthetic organisms, resulting in high chloropigment concentrations in surface waters (Supplementary Figure S1). Although epipelagia is truly oligotrophic, as defined by levels of inorganic nutrients, both the meso- and bathypelagia are characterized by high inorganic and low organic nutrient concentrations (Supplementary Table S1). The mesopelagia sample was taken at the oxygen minimum zone (Supplementary Figure S1), where pressure was approximately one-fifth of that at the point sampled in the bathypelagia.

Flow cytometry counts placed numbers of Prochlorococcus and heterotrophic Bacteria at the higher end of historical data covering 102 cruises (Supplementary Figures S2a and b), whereas Synechococcus and Eucarya abundances were at the lower end (Supplementary Figures S2c, d). Cell counts indicate that cyanobacteria (Prochlorococcus and Synechococcus) accounted for 30% of total prokaryote abundance at 5 m compared with an average of 25% during the 102 cruise data set.

Analysis of tag sequences

In total, 444 147 sequences passed our quality control methods, of which 66 990 were unique at a given depth (Table 2). Unique sequences and abundance data obtained from this study are available at http://www.ifa.hawaii.edu/UHNAI/Gayle/. A comparison of unique sequences at each depth revealed remarkably little overlap; 3707 sequences (5.9%) were detected at two depths, whereas only 298 (0.5%) were detected at all three depths. Total unique sequences from all depths numbered 62 975 and represented all domains (Table 2). A total of 6652 sequences (1.5% of those analyzed), including 3300 unique sequences, could not be assigned to any one domain by BLAST and were removed from further analyses.

Table 2 Domain level distribution of unique and total sequence abundances retrieved from samples in the epi- (10 m), meso- (800 m) and bathypelagic (4 400 m) zones of the oligotrophic open ocean at Station ALOHA
Full size table

Community structure

Although rarefaction analyses indicate an undersampling of the community in all three domains (Supplementary Figure S3), our large sequencing effort provided a substantial subsample from which to generate parametric (maximum likelihood) and nonparametric (ACE1) richness estimates and lower bounds (ACE and Chao1) at different levels of phylogenetic resolution (Supplementary Table S2). Richness estimates within the Archaea were lowest of the three domains, but higher at 800 m and 4400 m than at the surface (Figure 1a). Rank abundance and k-dominance plots for Archaea at all depths were similar, indicating that this domain displayed the least evenness in community composition, with sequence abundances of the top 10 unique tags contributing >60% to the total sequence count at all depths (Supplementary Figures S4, S5). We used a two-way clustering algorithm to cluster tags from two different depths at a variety of phylogenetic resolutions. This takes into account both tag identity and tag abundance and provides a measure of the proportion of the total number of tag sequences from one depth that have a taxonomic equivalent (at a specified cutoff) at the other depth. There was a greater taxonomic overlap between the Archaea at 800 and 4400 m than either had with the surface sample (Figure 2a). In fact, at 98% tag sequence identity, over 90% of the unique Archaea tags from the 800 m sample had a taxonomic equivalent in the 4400 m sample, and vice versa. However, even at 80% tag sequence identity, ∼5% of sequences from the 800 and 4400 m samples have no ‘relative’ in the surface sample.

Figure 1
figure 1

Observed and estimated richness of Archaea, Bacteria and Eucarya, calculated at a variety of phylogenetic levels. Left hand column (a) Archaea, middle column (b) Bacteria, right hand column (c) Eucarya.

Full size image
Figure 2
figure 2

(a) A two-way comparison of the percentage abundance of the Archaea sequence tags from each depth that cluster with tags from each of the other depths at different levels of phylogenetic similarity. The z axis represents the two tag sequence databases that are being compared. The x axis represents the level of tag sequence identity (1 (identical) to 0.8). The y axis represents the relative percentage of the tag sequences occurring in the second database from the z axis, which are phylogenetically ‘accounted for’ in the first database from the z axis. This analysis takes into account both tag identity and tag abundance and provides a measure of the proportion of the total number of tag sequences from one depth that has a taxonomic equivalent (at a specified cutoff) at the other depth compared. For example, consider a cluster of sequences (98% identical) at 10 m that accounts for 1.5% of the total number of sequences at that depth, and for which there is a taxonomically equivalent cluster (at 98% sequence identity level) at 800 m, which accounts for 0.01% of the total tag sequences at 800 m. This is reported in the two-way analysis as 1.5% of the tag sequences from 10 m being taxonomically ‘accounted for’ at 800 m, but only 0.01% of the tag sequences from 800 m being ‘accounted for’ at 10 m. (b) A two-way comparison of the percentage abundance of the Bacteria sequence tags from each depth that cluster with tags from each of the other depths at different levels of phylogenetic similarity. The z axis represents the two tag sequence databases being compared. The x axis represents the level of tag sequence identity (1 (identical) to 0.8). The y axis represents the relative percentage of the tag sequences occurring in the second database from the z axis, which are phylogenetically ‘accounted for’ in the first database from the z axis. (c) A two-way comparison of the percentage abundance of the Eucarya sequence tags from each depth that cluster with tags from each of other depths at different levels of phylogenetic similarity. The z axis represents the two tag sequence databases being compared. The x axis represents the level of tag sequence identity (1 (identical) to 0.8). The y axis represents the relative percentage of the tag sequences occurring in the second database from the z axis, which are phylogenetically ‘accounted for’ in the first database from the z axis.

Full size image

Within the Bacteria there was a transition in the trends in richness depending on phylogenetic resolution. In general, richness at the highest phylogenetic resolution, that is, ‘species/strain’ level (100% and 99% sequence identity), decreased with depth (Supplementary Table S2, Figure 1b). However, higher order phylogenetic richness estimates (for example, 95–80% identity) increased with depth (Figure 1b). Effectively, there was greater richness at the ‘phylum/class’ level with increasing depth. Bacteria composition at 10 m was less evenly distributed and showed greater dominance than at 800 or 4400 m (Supplementary Figures S4, S5). Two-way clustering of tag databases indicated that 800 and 4400 m samples hosted a similar overall taxonomic composition, although the relative abundances of taxa varied considerably (Figure 2b). Furthermore, although the 800 and 4400 m samples contained much of the phylogenetic diversity found at 10 m, the converse was not true. This may reflect an impact of sinking material on the composition of Bacteria in deeper waters, or alternatively for the presence of many closely related, but less abundant, deep phylotypes of common surface taxa. Comparison of Bacteria richness estimates indicates that, at the highest phylogenetic resolution, the samples analyzed here contain taxonomic diversity equivalent to that reported in North and South American soils similarly analyzed by V9 pyrosequencing (Supplementary Table S3). At all other levels of phylogenetic resolution, however, soil hosts greater microbial diversity.

Eucarya unique sequence richness peaked at 800 m (Figure 1c). However, at all other phylogenetic levels, richness decreased with depth (Figure 1c). Rank abundance and k-dominance plots were similar at all depths and indicated that Eucarya had the most even population distributions (Supplementary Figure S4). When tags from two depths were clustered together, the communities showed little overlap (Figure 2c) compared with that observed in the Bacteria and Archaea. Eucarya taxonomic assemblages thus changed significantly with depth.

Taxonomic transitions: Archaea

A total of 12 Archaea phylotypes were observed at all depths, whereas 49% of phylotypes were significantly overrepresented at a given depth. Overrepresentation may indicate that a phylotype is better adapted to conditions at a given depth than at the other depths analyzed. The dominant Archaea clade at 10 m was the Euryarchaeota Marine Group II (84.5% of Archaea tag sequence abundance at 10 m, 30.4% at 800 m and 37% at 4400 m) (Figure 3a, numerical abundances are provided in Supplementary Table S6). Abundance of Euryarchaeota Marine group III sequences was similar throughout the water column, whereas that of the Euryarchaeota Methanopyri increased with depth. The Crenarchaeota Marine group I, which includes the ammonia-oxidizing Candidatus Nitrosopumilus, the unclassified Crenarchaeota and the Cenarchaeum, accounted for 55.7% and 47.1% of Archaea tag sequences at 800 and 4440 m, respectively.

Figure 3
figure 3

(a) The abundance of major Archaea phylogenetic lineages observed at 10 m (outside ring), 800 m (middle ring) and 4 400 m (inside ring) depths. (b) The abundance of major Bacteria phylogenetic lineages observed at 10 m (outside ring), 800 m (middle ring) and 4 400 m (inside ring) depths. (c) The abundance of major Eucarya phylogenetic lineages observed at 10 m (outside ring), 800 m (middle ring) and 4 400 m (inside ring) depths.

Full size image

Taxonomic transitions: Bacteria

Of the 2978 Bacteria phylotypes observed from common high-scoring pair analysis, 320 were observed at all depths, whereas 19.8% were significantly overrepresented at a given depth. Tag sequences representing organisms from the Alpha- and Gamma Proteobacteria were abundant at all depths (Figure 3b, Supplementary Table S6), but at finer taxonomic resolution, the phylogenetic composition within these classes varied dramatically (Supplementary Figures S6, S7). Sequence tags that affiliated with the Alpha Proteobacteria, Cyanobacteria, Bacteroidetes and unclassified Greengenes OTU181, all decreased in relative abundance from 10 m with increasing depth (Figure 3b). Those sequences that increased in abundance at 800 or 4400 m included the Verrucomicrobia, Actinobacteria, Chloroflexi, Planktomycetaceae, Beta Proteobacteria, Acidobacteria and Firmicutes. Major groups of Bacteria that were not observed at 800 and/or 4400 m, but were observed at 10 m, include Candidate divisions BRC-1, OD1, OD3 and WS3, Aquifaceae, Deferribacteres, Deinococcus-Thermus, Fibrobacteres ktedobacter, Spirochaetes, Thermobaculum and the genus Mariprofundus in the Proteobacteria.

Taxonomic transitions: Eucarya

Of the 1313 Eucarya phylotypes observed by common high-scoring pair analysis, 230 were detected in all three samples, whereas 34.2% were significantly overrepresented at a given depth. Ribosomal tag sequences representing members of the Alveolata were the most abundant microbial eukaryote. However, the major subgroups within the Alveolata displayed different patterns of abundance with depth (Figure 3c, Supplementary Table S6); for example, Dinophyceae tag sequence abundance dropped markedly in the bathypelagia, whereas that of the marine Alveolate groups I and II (including the Amoebophyra) increased in the meso- and bathypelagia. An analysis of previously defined marine Alveolate groups I and II subclusters revealed that, although most were present at all depths, some were clearly more abundant in either surface or deeper waters (Supplementary Figure S8). Tag sequences associated with several radiolarian classes increased in relative abundance with depth, including the Acantharea (Taxipodida), RAD-III and the Sticholonchidae. Conversely, Polycystinea (Radiolaria) sequences, which were abundant at 10 m (16% of relative abundance) and peaked at 800 m (26%), were almost absent (0.2%) from 4400 m. Tag sequences affiliating with the marine Stramenopiles (MAST) and Haptophyceae, and with the novel Picobiliphyte cluster, decreased in abundance with depth.

Comparison with other marine samples

UPGMA clustering (see Supplementary materials for methods) of relative abundances of phylum level Bacteria lineages identified from 16S rDNA clone libraries from seven depths at Station ALOHA (DeLong et al., 2006) and from 16S rDNA tag sequencing of deep Labrador Seawater (Sogin et al., 2006), and 16S rDNA fragments parsed from metagenomic analysis of 43 surface ocean samples (Rusch et al., 2007) show that a higher order community structure within deep-sea assemblages is consistently different and forms a separate cluster from surface assemblages (Figure 4). A canonical analysis of principal coordinates analysis (see Supplementary materials for methods) further highlights the separation between surface and deep communities. Correlations of environmental variables indicated depth as the major contributing factor to this separation. The correlation of Bacteria lineages to canonical axes (Figure 4) highlights the major contributors to deep community separation and includes relative abundances of the Delta Proteobacteria, Chloroflexi, Gemmatomonadetes, Acidobacteria, Verrucomicrobia and Nitrospirae. The 10-m sample clusters within the major surface group, adjacent to other tropical open-ocean surface samples, which also included others from the photic zone at Station ALOHA.

Figure 4
figure 4

UPGMA clustering and canonical analysis of principal coordinates analysis of a higher order Bacteria community structure indicate a consistent shift from surface to deep waters. Samples in the analysis include those from the Global Ocean Survey (GOS, Rusch et al., 2007), Station ALOHA (ALOHA, Delong et al., 2006 and HOT169, this study) and the Labrador Sea (LS, Sogin et al., 2006). Asterisk (*) indicates samples from water >20 °C. Bacteria lineages affecting sample distribution in CAP analysis are coded as follows: D, deltaProteobacteria; G, Gemmatimonadetes; C, Chloroflexi; A, Acidobacteria; L, Lentisphaera; V, Verrucomicrobia; N, Nitrospirae; F, Firmicutes; G, gammaProteobacteria; Ap, alphaProteobacteria; B, Bacteroidetes; Cy, Cyanobacteria; S, Spirochetes; Bp, betaProteobacteria; Ac, Actinobacteria.

Full size image

Origin and novelty of rare and abundant phylotypes

The abundance of sequences from all domains related to those identified in deep-sea environments increased with the increasing depth. However, over half the rare Bacteria phylotypes observed at 800 and 4400 m, and over a quarter of those at 10 m, returned best matches to sequences in GenBank from non-pelagic sources, including soil and sediment (Figure 5a). Approximately one-quarter of the rare Eucarya phylotypes were similarly affiliated (Figure 5b). Rare phylotypes in all domains (rare Archaea considered as phylotypes with <10 sequences) were more likely to have a lower sequence similarity score (that is, novel) when compared with the Genbank database than were abundant ones (Figure 6). Bacteria phylotypes from 800 and 4400 m displayed greater novelty than those from 10 m, but the converse was true of Archaea phylotypes. Archaea sequences from 10 m displayed a peak in novelty from 95–97% sequence identity, which accounts for 49% of total abundances, providing evidence of a relatively abundant surface-associated Archaea group (∼genera level) not reported in previous molecular surveys. Eucarya phylotypes were as likely to be novel, regardless of the depth from which they were sampled. On an average, 32% of abundant and 44% of rare Eucarya phylotypes displayed <95% identity with their best BLAST matches in the GenBank database, compared with 5% and 11% for Archaea and 3% and 13% for Bacteria. Furthermore, 62% of sequences that did not significantly match GenBank sequences (and so were removed from the above analysis) were >90 bp in length; this, when over 99% of all sequences identified as Bacteria or Archaea were 90 bp or less, making it more likely that sequences not matching any others in GenBank, if ribosomal in origin, are eukaryotic.

Figure 5
figure 5

(a) The environment of origin of the highest-scoring pairs in the Genbank database for rare Bacteria tag sequences occurring at 10 m (outside ring), 800 m (middle ring) and 4400 m (inside ring) depths. (b) The environment of origin of the highest-scoring pairs in the Genbank database for Eucarya tag sequences occurring at 10 m (outside ring), 800 m (middle ring) and 4400 m (inside ring) depths.

Full size image
Figure 6
figure 6

The average percentage identity of ribosomal tag sequences to GenBank HSPs for three depths (10, 800 and 4400 m) at Station ALOHA.

Full size image

Discussion

Novel technologies continue to expand our understanding of microbial diversity and community structure. Metagenomic analysis (Venter et al., 2004; DeLong et al., 2006) and ribosomal tag pyrosequencing strategies (Sogin et al., 2006; Huber et al., 2007) have previously identified ‘unexpectedly’ high Bacteria and Archaea phylogenetic and functional diversity. Here, we extend this observation to the microbial Eucarya. Even at the conservative 95% tag sequence identity (∼97% SSU rRNA similarity), our OTU observations and richness estimates are almost one order of magnitude higher than any previously reported (Supplementary Table S2) (Jeon et al., 2006; Zuendorf et al., 2006; Countway et al., 2007). The difference between our Eucarya richness estimates for unique and clustered sequences indicates that Eucarya display equivalent levels of microdiversity to the Bacteria community (Acinas et al., 2004; Brown and Fuhrman, 2005). For many clades, this microdiversity may represent population, rather than intra-genomic, diversity, given the fact that not a single polymorphism was found in a tandem repeat array of the rDNA operon from a fosmid representing an organism within the Novel Avleolate group I, cluster I (NA1.1) (Massana et al., 2008). The Novel Alveolate groups I and II (Groisillier et al., 2006) collectively contributed to 12.4, 23.6 and 21.2% of tag sequence abundance at 10, 800 and 4400 m, respectively. Recently, sequences clustering within these clades have been identified as representing endoparasitic dinoflagellates related to the genera Amoebophyra (Kim et al., 2008) and Duboscquella (Harada et al., 2007). Infection by both genera is mediated by the production of dispersal spores and infections and is mostly lethal (Kim et al., 2008). If the diversity within NAI and NAII does indeed represent organisms displaying a potentially lethal parasitic lifestyle, this may have significant impacts on our concepts of the mechanisms controlling host population density and diversity. The various subgroups within NAI and NAII varied with depth (Supplementary Figure S8). These patterns of (putative) parasitic abundance and diversity may be associated with either host or parasite characteristics. Members of the Amoebophyra and Dubosquella are known to parasitize a range of organisms including dinoflagellates, radiolarians, ciliates, acanthareans and sticholonche (Kim et al., 2008), all of which displayed different patterns of diversity and abundance in our depth gradient. Multiple species of Amoebophyra have been reported to infect a single host species (Salomon et al., 2003), but strong host specificity may drive the genetic diversification of the host species. Conversely, lethal infections may reduce the dominance of superior competitors, allowing the coexistence of many competing species, analogous to the viral ‘kill the winner’ hypothesis, previously shown to alter community structure and increase diversity in bacterial populations (Fuhrman and Schwalbach, 2003). These effects may go some way toward explaining the diversity in eukaryotic populations. Although some of the observed host diversity may be environmentally neutral, it may also act to boost a population's biological robustness (which enables a system to maintain its functionalities against external and internal perturbation) (Kitano, 2004) by confounding parasitic attack. A similar mechanism has been proposed for bacterial microdiversity (Brown and Fuhrman, 2005). Furthermore, the mass release of spores from an infected or dead host would have an effect in returning significant amounts of carbon from sinking material to the immediate environment, enhancing carbon flow through the microbial loop, as has been recently implied for freshwater systems (Lefèvre et al., 2008).

Besides the persistent presence of NAI and NAII, some general phylogenetic patterns have emerged from the few molecular surveys of Eucarya carried out at various locations (Epstein and López-García, 2008) across depth profiles. Both Countway et al. (2007) and Not et al. (2007) reported shifts in community structure when comparing euphotic zones with deeper waters in the North Atlantic and Sargasso Sea, respectively. As with this analysis, Countway et al. (2007) observed decreasing OTU abundance with depth (although only one of three shallow versus deep comparisons showed statistically significant differences). A potential explanation for this is that decreased resource availability (Karner et al., 2001) leads to reduced niche partitioning in deeper waters. Further analysis is required to determine whether these changes are congruent across ocean basins. Much of the apparent Eucarya diversity is yet to be adequately described, as indicated by the degree of novelty shown by Eucarya tag sequences compared with those of the Bacteria and Archaea (Figure 6).

The 800 and 4400 m samples were more closely related phylogenetically than was either to the 10 m sample for all domains, despite the five times greater physical distance between these sites. A number of factors likely contribute to this. Physicochemical conditions such as oxycline, chemocline and thermocline may provide a strong barrier to genetic exchange/community homogenization between the surface and the two deeper sample sites. Furthermore, secondary adaptation to pressure may require relatively little genetic change for low-temperature-adapted Bacteria (Lauro and Bartlett, 2008), allowing organisms adapted to the temperature regime at 800 m to populate the higher pressure environment at 4400 m, which is only a few degrees cooler. This is consistent with the hypothesis that, in the absence of sunlight, temperature is a major ecological/taxonomic stratifying factor in the marine water column, overriding pressure at least at depths of up to 4000 m (Martín-Cuadrado et al., 2007). Indeed, UPGMA cluster analysis of phylum and class level Bacteria lineages from our study and others indicates a consistent shift in the higher order taxonomic structure in meso- and bathypelagia when compared with that in coastal and open-ocean surface samples (Figure 4). Geography and temporal variation may have an important role in the makeup of the deep-sea community structure, as evidenced by the clustering patterns of samples taken from the Labrador Sea (Sogin et al., 2006) and Station ALOHA in different years (this study; DeLong et al. (2006)). This observation agrees with studies that used other molecular methods to describe spatial heterogeneity in the bacterioplankton community composition in samples from marine epi-, meso- and bathypelagia (Lee and Fuhrman, 1991; Hewson et al., 2006a). However, some of the variation may be due to differences in methodology. The greater disparity in phylogenetic composition within the Eucarya between depths may mean that the physiological adaptations required for Eucarya survival in the oxygen-depleted deep sea may not be so easily expressed and may require greater genomic (and hence phylogenetic) differentiation.

A recent meta-analysis (Quince et al., 2008) comparing (meta)genomic data from marine surface samples with tag sequence data from different soils and deep-sea sediment habitats concluded that deep-sea diversity, at least in sediments near hydrothermal vents, and soil diversity significantly exceed diversity in the ocean surface. Our results, although from a single oligotrophic open-ocean site, but generated using consistent methodology across the depth profile, suggest that species/strain level diversity within the Bacteria is greater at the surface than at the depth in pelagia, whereas the reverse is true for a higher order phylum/class level. A confounding factor not considered by Quince et al. (2008) is the variability inherent in the different SSU rDNA helix regions examined. A comparison of the V6 and V9 regions excised from near full-length Bacteria 16S sequences indicates that the V6 region examined by Sogin et al. (2006) and by Huber et al. (2007) is approximately twice as variable as the V9 region used in this study and by Roesch et al. (2007) to examine soil diversity (y=2.28, data not shown). Hence, comparisons using a simple 97% tag sequence identity cutoff when considering surveys using different variable regions should be avoided, and careful consideration should be paid to relative rates of divergence. Roesch et al. (2007) suggested that for the V9 region, 100% sequence similarity best equates to species level. Comparison of ACE and Chao1 statistics of our samples with those of the soil samples analyzed by Roesch et al. (2007) (re-analyzed by Quince et al. (2008)) reveals that, at this phylogenetic level, microbial diversity in the oligotrophic open ocean is equivalent to that in three agricultural soils and one boreal forest soil (Supplementary Table S3). At all other phylogenetic levels, however, the soil samples host two- to fourfold greater richness, clearly far from the orders of magnitude previously speculated to be the difference between aquatic and soil/sediment microbial diversity (Torsvik et al., 2002). Indeed, the difference may simply be a function of spatially constrained ecosystem processes in soil as opposed to those occurring throughout the integrated 4500 m water column in the open ocean. An interesting observation of our work, and that of Roesch et al. (2007), is that some samples seem to contain higher than anticipated phylum level diversity; for example, a Canadian boreal soil and our 4400 m sample. This may relate directly to a greater variety in functional pathways. At the surface, a higher Bacteria and Eucarya species/strain diversity probably reflects greater niche availability. The availability of photosynthetically active radiation provides an opportunity for mixotrophy, and drives photoautotrophy, the net result being increased concentrations, diversity and availability of dissolved organic matter in the euphotic zone. Such an enhanced dissolved organic matter pool may support many related heterotrophic lineages. Below 100 m, DOC concentrations decrease rapidly and dissolved organic matter in the deep ocean largely comprises the refractory pool that is not easily broken down by the microbial community (Carlson et al., 1994; Witte et al., 2003; Hopkinson and Vallino, 2005). It may be that those microbes that are more abundant in deep waters, yet are not part of shallow-water communities, have important roles in the degradation of the ‘refractory’ dissolved organic matter pool; the alternative metabolic pathways required may well explain the higher phylum level diversity observed in the deep sea. In this respect, the closest matches in the GenBank of many phylotypes rare in the meso- and bathypelagia were from non-planktonic habitats. Some of these matches may be due to allochthonous microbes, such as those from surfaces, digestive tracts, fecal pellets and so on, of higher organisms, or from suspended sediments and coastal erosion, aeolian processes or other processes. Indeed, many Bacteria recovered from deep-sea environments show no evidence of a physiological adaptation to in situ environmental conditions (Yayanos, 1986; Lauro et al., 2004), and survive, but do not grow, under prevailing low-temperature high-pressure conditions (Lauro et al., 2004). However, it is likely that the majority of organisms represented by rare tag sequences are, indeed, native to the ocean, but remains poorly represented in databases due to a lack of breadth and scale in sampling efforts. If this is indeed the case, microbes of the rare biosphere, which have the closest known homology to organisms from soil and sediment, may be responsible for key metabolic processes in global biogeochemical cycles, similar to those carried out by these related organisms in other environments.