Introduction

Over a decade of research has highlighted the critical importance of open ocean nitrogen fixation in supporting oligotrophic microbial food web N demands (Carpenter, 1983; Carpenter and Romans, 1991; Michaels et al., 1996; Karl et al., 1997; Capone, 2000; Codispoti et al., 2001). Nitrogen fixation is carried out by a diverse suite of microorganisms, including unicellular cyanobacteria, other Bacteria, Archaea, and colonial cyanobacteria (Dugdale et al., 1961; Carpenter, 1983; Villareal, 1990; Zehr et al., 2003). Filamentous, nonheterocystous cyanobacteria of the genus Trichodesmium are biogeochemically important organisms in tropical waters, fixing at least 80 Gt of N per year, and are globally distributed in warm tropical and subtropical waters (Capone et al., 1997; Capone and Carpenter, 1999; Subramaniam et al., 1999; Orcutt et al., 2001; Westberry and Siegel, 2006). Filaments (trichomes) of the cyanobacterium form aggregates that are arranged as puffs loosely connected at the center of filaments, or as polarized rafts, both of which form dense aggregations at the surface during calm conditions (Bowman and Lancaster, 1967; Capone et al., 1997). Blooms of Trichodesmium can be observed from space (Capone et al., 1997; Subramaniam et al., 1999; Westberry and Siegel, 2006), and represent a pseudobenthic surface on which other organisms may recruit (O’Neil and Roman, 1994; O’Neil et al., 1996). Despite their global importance, studies of Trichodesmium gene expression have been restricted to a handful of genes involved in heterocyst differentiation, global nitrogen regulation, glutamine synthesis, nitrogen fixation, iron stress and phosphorus acquisition (Zehr and McReynolds, 1989; Kramer et al., 1996; Webb et al., 2001; Dyhrman et al., 2002; El-Shehawy et al., 2003; Dyhrman et al., 2006).

Previous reports of the microflora associated with Trichodesmium colonies indicated that a wide spectrum of microorganisms are found closely associated with colonies, including viruses, bacteria, eukaryotic microorganisms and metazoa (Paerl et al., 1989; Siddiqui et al., 1992; Zehr, 1995; Ohki, 1999; Sheridan et al., 2002). The abundances of closely associated organisms are elevated compared with surrounding seawater (Sheridan et al., 2002), and the hydrolytic enzyme activities within the colonies are higher than outside colonies (Nausch, 1996). Therefore, colonies of Trichodesmium represent hotspots of biological activity in the low productivity and nutrient-poor waters where they inhabit. As a large amount of fixed N is released from the cells into the surrounding waters (Capone et al., 1994; Mulholland et al., 2006), co-occurring organisms can potentially benefit from localized N-enrichment surrounding and within the colonies (Capone et al., 1994). The enhanced N conditions within colonies may cause P limitation within the colony infrastructure and selection for organisms capable of enhanced P uptake. However, knowledge of the function and physiology of organisms inhabiting Trichodesmium colonies, especially under bloom conditions, is unknown.

Recent application of whole-genome random transcript sequencing (metatranscriptomics) has shown utility for understanding coastal and open ocean microbial community ecophysiology (Poretsky et al., 2005, 2009b; Frias-Lopez et al., 2008; Gilbert et al., 2008; Hewson et al., 2009). Unlike metagenomic surveys which elucidate potential genetic capabilities (Tyson et al., 2004; Venter et al., 2004; Rusch et al., 2007), and which are mostly focused on the discovery of novel metabolic pathways or extent of diversity, metatranscriptomics provide information on active processes in dominant microorganisms (Poretsky et al., 2005, 2009b; Frias-Lopez et al., 2008). Metatranscriptomic studies to date have elucidated that a large proportion of genes expressed in open ocean assemblages have no known function, or show less similarity to cultivated microorganisms than metagenomic surveys (Frias-Lopez et al., 2008; Gilbert et al., 2008). There are no published studies of community gene expression within particles, despite their importance as ‘hotspots’ of biological activity in the oligotrophic ocean.

The aim of this study was to examine the in situ gene expression of microbial assemblages associated with Trichodesmium during a bloom to determine the composition of active components of the associated microflora and to identify major metabolic characteristics of tightly associated microorganisms. We applied a metatranscriptomic approach to independent samples of Trichodesmium aggregates collected in the day and night. These results provide new information on the diversity and ecophysiology of microorganisms closely associated with Trichodesmium blooms.

Materials and methods

Sampling of Trichodesmium bloom sample

Samples for metatranscriptomic analysis were collected on board the R/V Kilo Moana during an intense bloom of the cyanobacterium at station KM070324 (15°S, 178°45′E) north of the Fiji Islands on 12 April 2007. The sea state at the time of sampling was calm (glassy surface) and Trichodesmium colonies were observed at the surface in a thin layer. Trichodesmium was collected using a 64-μm mesh plankton net. Immediately after the net was retrieved, the tow material was placed in a seawater-rinsed bucket. Subsamples of Trichodesmium were then collected within 5 min by skimming the floating material (Trichodesmium and associated organisms) into 50-ml centrifuge tubes which were immediately frozen in liquid nitrogen. The night sample was collected at 0100 h, whereas the day sample was collected at 0900 h. Samples for enumeration of viruses and bacteria were collected at 5 m using a Niskin bottle mounted on a CTD rosette. Seawater from the Niskin was retrieved into 50-ml centrifuge tubes to which 2 ml of 0.02-μm filtered formaldehyde was added, and samples processed immediately following an established protocol (Patel et al., 2007). Samples were transported on liquid nitrogen to the University of California Santa Cruz for analysis.

RNA extraction

The Trichodesmium samples were thawed in the laboratory by centrifuging tubes at 5000 × g for 20 min, which pelleted the Trichodesmium and associated microorganisms. The supernatant was decanted and cells (approximate volume was 5 ml, containing ∼5 × 106 Trichodesmium cells) were immediately placed on ice. Subsamples (∼0.1 ml) of cell pellets were removed using a pipette tip, placed into RNase-free 2-ml cryovials, then subjected to the RNeasy Plant Mini kit (Qiagen, Valencia, CA, USA) with the following modification. Glass beads (100 μl) and 450 μl of buffer RLT containing 1% β-mercaptoethanol were added, and the tubes placed in a bead beater for 2 min. After mechanical lysis, the homogenized material was processed according to the Qiagen protocol. The resulting RNA was eluted in deionized water and subsequently treated to remove DNA using the RNase-free DNase kit (Zymo Research, Orange, CA, USA).

mRNA enrichment

The total RNA sample containing rRNA and mRNA was subjected to two protocols to enrich the fraction of mRNA relative to rRNA following an approach applied previously to open ocean microbial communities (Poretsky et al., 2009a, 2009b). RNA was first subjected to terminator exonuclease treatment (which removes 5′-monophosphate-capped RNA) using the mRNA-ONLY protocol (Epicentre, Madison, WI, USA). rRNA was further reduced by subtractive hybridization using the MicrobExpress kit (Ambion, Austin, TX, USA) following the manufacturer's protocols. The resulting mRNA enrichment was purified (according to the MicrobExpress protocol) by precipitating, washing and resuspending the RNA in 20 μl of nuclease-free deionized water.

RNA in vitro amplification

The mRNA-enriched samples were amplified using in vitro transcription after mRNAs were polyadenylated, as part of the MessageAmp II—Bacteria aRNA kit (Ambion). The polyadenylation step was performed on 180 ng of the mRNA-enriched samples. The protocol does not select for prokaryotic mRNAs as all mRNAs are polyadenylated (even those which are already polyadenylated, that is, eukaryotic mRNA); hence the resulting aRNA contains eukaryotic, prokaryotic and phage transcripts. The aRNA was prepared according to the manufacturer's protocols, and eluted in a final volume of 150 μl of deionized water. Samples were subsequently concentrated in a speed evaporator (Thermo Savant, Waltham, MA, USA) to a volume of 50 μl and quantified using a spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA).

Double-stranded cDNA synthesis

The aRNA (15 μg) was converted to double-stranded cDNA (ds cDNA) by reverse transcription and second strand synthesis reactions. First, the aRNA was diluted to 500 ng μl−1 (10 μl total). Triplicate samples were then treated with 400 U reverse transcriptase (Superscript III; Invitrogen, Carlsbad, CA, USA), 10 nmol dNTPs (Invitrogen), 1 × First Strand Buffer (Invitrogen), 0.1 μmol dithiothreitol (DTT) and 500 ng random primers (Promega, Madison, WI, USA). The reactions were first heated to 70 °C for 10 min in the presence of the random primers and dNTPs, then cooled on ice before addition of the remaining reagents. The reactions were then incubated at 50 °C for 50 min. After first strand synthesis, 1 × Second Strand Buffer (Invitrogen), 6 μmol dNTPs, 10 U E. coli DNA ligase, 40 U E. coli DNA polymerase and 2 U RNAse H were added to the reactions, which were then incubated at 16 °C for 2 h. At the conclusion of the reaction, each tube was amended with 10 U of T4 DNA ligase and incubated at 14 °C for 10 min. Following the second strand synthesis, the samples were treated with 20 μg of RNAse A at 37 °C for 30 min. RNAse A treatment was terminated by mixing the samples with 50% phenol:chloroform:isoamyl alcohol (24:1:0.1) by inversion, after which the aqueous layer was removed. Nucleic acids in the aqueous layer were precipitated with 16 μl of 7.5 M NH4COOH, 35 μg Glycogen (Ambion) and 326 μl of 100% EtOH, and precipitated overnight at −20 °C. The samples were then centrifuged at 15,000 × g for 1 h. Supernatant was decanted and pellets were rinsed with 70% EtOH at −20 °C. The pellets were dried in a speed evaporator (Savant) for 10 min, resuspended in 20 μl of distilled H2O, and replicate reactions were combined. Small ds cDNA fragments and other materials were removed by passing the ds cDNA through a cleanup kit (Zymo Clean & Concentrator −5). The cleaned ds cDNA was quantified by spectrophotometer at 260 nm, and size range and concentration of fragments were determined by running 1 μl on an Agilent (Santa Clara, CA, USA) Bioanalyzer DNA 1000 chip. The samples were then pyrosequenced as described elsewhere in picoliter reactors on a GS FLX platform (454 Life Sciences, Branford, CT, USA) (Margulies et al., 2005). Sequence reads from the two metatranscriptomes are available at the Community Cyberinfrastructure of Advanced Marine Microbial Ecology Research and Analysis under accession CAM_P0000051.

Bioinformatic analysis

The pyrosequence reads were initially analyzed to remove replicate sequences by comparing the first 100 bp of sequence in Microsoft Excel. As random primers were used in the creation of cDNA libraries, and furthermore the cDNA was sheared before sequencing, it is unlikely that replicate sequences represent true replicate transcripts, but rather are an artifact of the sequencing protocol. Furthermore, poor-quality sequence towards the end of the sequence read length precludes removal of replicate sequence over the entire read length. After replicate sequences were removed (leaving one sequence to represent each replicate), the dereplicated libraries were compared against the Ribosomal Database Project (RDP II; (Cole et al., 2007)), a boutique database of 23S rRNAs and a boutique database containing the 5.8S, 18S and 28S of common marine eukaryotic microorganisms, using BLASTn (Altschul et al., 1997). Sequences matching at E-values of <10−3 were discarded from the dereplicated sequence library as they represented microbial rRNAs. In addition, sequences <75 bp or containing >60% of any single base were discarded.

The mRNA libraries were compared by BLASTx against the All Prokaryotic Proteins, All Eukaryotic Microbial Proteins and All Viral Proteins databases in the Community Cyberinfrastructure for Advanced Microbial Ecology Research and Analysis (CAMERA) server (http://camera.calit2.net/) (Rusch et al., 2007). The putative taxonomic affiliation and function were assigned as the top match for each read which had an E-value of <10−3. Sequence reads with no matches in the three protein databases were compared by BLASTn analysis against the non-redundant (nr) database at NCBI and all assembled metagenomic reads of the Global Ocean Survey (GOS) to further resolve hits. These sequence reads were compared by BLASTx against all assembled proteins at CAMERA, and the functional category of protein matches was determined by comparing the sequence reads against the Kyoto Encyclopedia for Genes and Genomes (KEGG) database. Furthermore, remaining unclassified sequences were compared with the nr protein sequence database at NCBI using BLASTx. Prokaryotic and microbial eukaryotic reads with E-values <10−3 were compiled with the previous hits. Data from additional remaining reads with bit score ⩾40 (E-value ∼0.06) were considered separately, as this has been the lower cutoff value used in previous studies (Frias-Lopez et al., 2008; John et al., 2009).

Results and Discussion

Microscopic observations of Trichodesmium bloom samples

Trichodesmium abundance in surface waters was ∼1.13 × 104 trichomes per liter and was primarily of raft morphology with a large number of free filaments. The abundance of heterotrophic bacteria immediately below the bloom at 5 m, as measured by SYBR Green I staining and epifluorescence microscopy (Noble and Fuhrman, 1998; Patel et al., 2007), was 2.8 × 106 cells per ml and virus abundance of 2.4 × 107 viruses per ml (I Hewson, unpublished data). The abundance of Synechococcus and Prochlorococcus at 15 m, as measured by phycoerythrin autofluorescence flow cytometry, was 6 × 103 cells per ml and 3 × 101 cells per ml, respectively (B Carter, unpublished data).

Owing to the extremely heterogeneous nature of Trichodesmium colonies, it was not possible to quantify the associated organisms accurately, but we qualitatively describe the biomass of co-occurring microorganisms observed at night in a SYBR Green I-stained sample prepared from the bloom depth and from light microscopic observations of bloom samples. In SYBR Green I-stained samples, we observed bacteria attached to single Trichodesmium trichomes and high concentrations of long filaments between trichomes in colonies of Trichodesmium (Figure 1). We also observed cells resembling Synechococcus and non-phycoerythrin-containing cells within the colonies. Under light microscopic examination, we observed multiple morphologies of Trichodesmium and other cyanobacteria, including Phormidium-like filaments. Morphologically diverse eukaryotic flora were also seen among the Trichodesmium colonies, including dinoflagellates, diatoms, ciliates, silicoflagellates, a foram, and radiolarians (Figure 2). In addition, at least one unidentified metazoan was observed within raft colonies of Trichodesmium.

Figure 1
figure 1

Photomicrographs of SYBR Green I-stained samples of Trichodesmium colonies at night. The filaments (A) forming rafts (likely Trichodesmium erythraeum) had a large number of filamentous bacteria (B) intertwined with filaments, whereas other filaments had individual bacterial cells (C) attached to the outside of the filaments. Scale bar=10 μm.

Full size image
Figure 2
figure 2

Photograph of the Trichodesmium bloom at KM070324 (a) and photomicrographs of organisms associated with Trichodesmium colonies at night: (b) copepod nauplii larva , (c) harpactacoid copepod, (d) radiolarian, (e) diatoms and dinoflagellates, (f) Lyngbya-like cyanobacterium among Trichodesmium filaments, (g) Foraminifera and (h) unidentified coccoid microorganisms. Scale bar=50 μm.

Full size image

Metatranscriptome library characteristics

We obtained a total of 5711 sequences in the day metatranscriptome and 5385 sequences in the night metatranscriptome that were putatively mRNAs >75 bp and with individual base contents <60% of total metatranscript sequence (Table 1). The mean sequence length was 210 bp in the day and 213 bp in the night. The mean G+C content of metatranscripts was 50.9±0.1% in the day and 51.6±0.1% in the night. The number of sequence reads matching prokaryotic, eukaryotic and viral databases at CAMERA at E-values <0.001 and Genbank nr at bit score ⩾40 (which for sequence alignments in our study equates to an E-value of ∼0.06) varied in the day and night (Table 1).

Table 1 Summary of number of sequence reads matching proteins and metagenomes retrieved from Trichodesmium metatranscriptomes in the day and night
Full size table

The majority of matches to proteins were at E-values >10−20 (63 and 57% of all hits in the day and night, respectively). Previous metagenomic and metatranscriptomic studies have used various E-value cutoffs to annotate libraries of sequences with approximately the same length as those in this study, including ⩽10−4 (Thurber et al., 2008), ⩽10−3 (Culley et al., 2006; Dinsdale et al., 2008; Gilbert et al., 2008), ⩽10−2 (Poretsky et al., 2009a, 2009b), and bit score cutoff of ⩾40 (Frias-Lopez et al., 2008). As the E-value of read is related to the length of alignment (shorter read matches have higher E-values; Wommack et al., 2008), we chose to use an E-value cutoff of E<0.001, and separately analyze reads matching at bit score ⩾40 in line with previous metatranscriptomic efforts. Furthermore, as we focused our study on the frequency of individual genes among the randomly sampled transcript pool, we chose not to assemble the metatranscriptomic sequence data.

The taxonomic affiliation of bacterial mRNAs was different between the two metatranscriptomes (Figure 3a). Trichodesmium mRNAs made up the dominant fraction of all annotated mRNAs in the day sample (30%), but made up only a small fraction of all night prokaryotic reads (4%). Eukaryotic microbial mRNAs (Figure 3b) were primarily most closely related to proteins in Fungi (30 and 27% of all recognized eukaryotic mRNAs in day and night, respectively), followed by Alveolates (25% day and 22% night), Viridiplantae (11% day and 7% night) and Stramenopiles (10% day and 7% night). Viral mRNAs in the day and night metatranscriptomes were primarily cyanophage, with fewer eukaryotic virus and bacteriophage mRNAs. Lower stringency (bit score ⩾40) results from comparisons against the nr database were heavily affiliated with metazoans (80% day and 90% night; Table 2).

Figure 3
figure 3

Taxonomic affiliation of (a) prokaryotic and (b) eukaryotic Trichodesmium colony-associated putative mRNAs as a percentage of each group. Taxonomic affiliation was assigned as the best BLASTx hit of mRNAs to the All Prokaryotic Proteins and All Eukaryotic Proteins at CAMERA, with E-values <10−3. Where mRNAs matched both prokaryotic and eukaryotic proteins, the lowest E-value match was assigned.

Full size image
Table 2 Top 10 most hit proteins to the non-redundant protein database in both the day and night samples, using the sequences not matching annotated proteins or rRNAs at E <0.001
Full size table

Our results highlight the richness of microorganisms inhabiting Trichodesmium colonies under bloom conditions. Previous studies have observed the presence and microbial activities of bacteria on the surface of Trichodesmium colonies (Paerl et al., 1989; Zehr, 1995; Nausch, 1996; Dyhrman et al., 2002). However, there have been few studies documenting the composition of bacteria inhabiting the Trichodesmium colonies. The large number of γ-proteobacterial transcripts, especially from Alteromonas, Marinobacter and Pseudoalteromonas, as well as from members of the Bacteroidetes suggests that Trichodesmium is colonized by fast-growing and opportunistic microorganisms, and composition of assemblages is different from free-living bacterioplankton communities (Fuhrman and Ouverney, 1998; Giovannoni and Rappe, 2000; Venter et al., 2004; Rusch et al., 2007). In a study of coral mucus additions to seawater (Allers et al., 2008), a rapid increase in the abundance of Alteromonas bacteria concomitant with significant decreases in the C:N ratio of particulate organic matter was found, suggesting that they rapidly assimilated dissolved organic carbon. Marinobacter is the most common heterotrophic organism associated with libraries of the assimilatory nitrate uptake gene nasB in nitrate-rich waters (Allen et al., 2001). Trichodesmium is known to exude fixed N and C into surrounding waters (Capone et al., 1994; Mulholland et al., 2006). The Bacteroidetes group includes those which are surface-associated in the marine environment and those which are capable of degrading macromolecules by hydrolytic enzyme production (Giovannoni and Rappe, 2000). Within the trichosphere, it is hypothesized that organic C and N concentrations are several fold higher than the surrounding seawater, which is reflected in higher heterotrophic activity and increased hydrolysis of several compounds within colonies relative to surrounding waters (Nausch, 1996). This environment may select for fast growing opportunistic taxa. Our results are consistent with a previous study using fluorescent in situ hybridization on Trichodesmium-associated bacteria, which found large numbers of Flavobacteria and enteric bacteria (presumably γ-proteobacteria) (Zehr, 1995). We also observed α-proteobacterial mRNAs, which is in contrast to observations of Trichodesmium colonies reported by Zehr, 1995.

Transcripts from cyanobacteria other than Trichodesmium included those matching Synechococcus, Prochlorococcus, Crocosphaera, Nostoc and Lyngbya proteins. Prochlorococcus, Crocosphaera and Synechococcus transcripts may have been the contaminants from the surrounding seawater, which contained abundant picocyanobacteria (unpublished data) or cyanobacteria attached to the Trichodesmium colonies. Cyanobacteria resembling Synechococcus and Phormidium were observed microscopically within bloom samples (Figure 2), consistent with previous observations of these genera within Trichodesmium colonies (Siddiqui et al., 1992; Sheridan et al., 2002).

Our observations of microbial eukaryotic mRNA taxonomic affiliation are consistent with microscopic observations, which showed the presence of morphologically diverse eukaryotic flora among the Trichodesmium colonies (Figure 2), as well as previous studies demonstrating a diverse eukaryotic flora and fauna associated with Trichodesmium (Sheridan et al., 2002). The large number of putative fungal transcripts in the mRNA libraries is consistent with previous observations of fungi in Trichodesmium colonies (Sheridan et al., 2002). However, it may also reflect greater representation of fungal genomes in the eukaryotic microbial proteins database. Diatoms were also observed colonizing Trichodesmium, where it was speculated that the benthic-like environment allowed heavily silicified diatom taxa to exist (Sheridan et al., 2002).

The phage and viral sequences recovered may have originated in either Trichodesmium or the associated organisms. The large number of cyanophage-like sequences in the day, when Trichodesmium comprised the greatest proportion of total mRNAs, suggests that some of the viral sequences may be associated with Trichodesmium. Previous reports of virus-like particles in dying and mitomycin C-treated Trichodesmium cells (Ohki, 1999), and the production of virus-like particles from filaments (Hewson et al., 2004) suggest that the cyanobacterium may be affected by viruses in nature. However, no genomic information on potential Trichodesmium cyanophages is available. The viral genes detected in this study may be attractive targets for future study of virus–Trichodesmium interactions.

Transcript library gene orthologs

The detected transcripts of Trichodesmium show the key importance of nitrogen fixation and photosynthesis in the ecology of the microorganism under bloom conditions, whereas heterotrophic bacteria were primarily involved in growth and energy metabolism. Of the 198-day and 19-night Trichodesmium mRNAs (Figure 4), the most common orthologs were hypothetical proteins or proteins of unknown function (31 in the day and eight at night), transcripts related to the photosynthetic apparatus (28 in the day and two at night) or involved in nitrogenase enzyme activity (18 in the day and one at night; Figure 4). Prokaryotic genes, which did not match Trichodesmium proteins (Figure 4), were primarily hypothetical proteins or proteins of unknown function. Among the 449-day and 508-night transcripts, 101 during the day and 96 at night were conserved hypothetical proteins or proteins of unknown function, whereas 53 in the day and 83 at night were either ribosome components or genes involved in DNA replication and repair. The remaining mRNAs were mostly present as singletons in the libraries and involved in photosystem apparatus biosynthesis, oxidative phosphorylation, and transport, P acquisition, signal processing, and As detoxification.

Figure 4
figure 4

Comparison of the orthology of total number of metatranscripts between day and night assigned to Trichodesmium and prokaryotic organisms other than Trichodesmium.

Full size image

Our results show that the dominant gene expression of Trichodesmium colonies is similar to free-living pelagic marine microorganisms (Frias-Lopez et al., 2008; Hewson et al., 2009; Poretsky et al., 2009b), with heavy genetic machinery investment in energy metabolism and growth. Trichodesmium colony metatranscripts were mostly associated with ribosome synthesis, RNA polymerase and other replication and repair enzymes (14% of all metatranscripts). Transcripts of genes involved in energy metabolism, including photosynthesis, oxygenic phosphorylation and the citrate cycle also comprised a large component of total Trichodesmium colony metatranscripts (15%). These results show that energy acquisition and growth were critical to Trichodesmium and other closely associated organisms within the bloom.

Trichodesmium transcripts included those implicated in sulfur metabolism, including methionine synthase and adenosylhomocysteinase, both enzymes involved in the S-adenosyl methionine (SAM) cycle, used in transmethylation reactions in proteins, nucleic acids and lipids (Koshiishi et al., 2001). Trichodesmium has extensively methylated adenine in its genomic DNA (15 mol%) (Zehr et al., 1991). Trichodesmium transcripts also included the arsA component of the arsA/B-encoded arsenite efflux pump, and thus a potentially active pathway for arsenate detoxification in this bloom population (see below).

The large number of Trichodesmium transcripts associated with nitrogenase activity (8% of all transcrtipts), including nifH, nifD and nifE confirms previous studies demonstrating the high expression of nitrogenase relative to other genes of diazotrophs (Stoeckel et al., 2008), and indicates that it is an active biosynthetic pathway. It is interesting to note that no nif genes were recovered from organisms other than Trichodesmium. Two transposases were detected among the 198 Trichodesmium metatranscripts. The function of transposases in open ocean cyanobacteria is unknown, however in other related diazotrophic cyanobacteria, tranposases are present in high genome copy number and recently have been observed as dominant transcripts in situ (Hewson et al., 2009). The large number of conserved hypothetical proteins or proteins of unknown function (31 in the day and eight at night) is consistent with observations in other metatranscriptomic surveys (Poretsky et al., 2005, 2009b; Frias-Lopez et al., 2008; Gilbert et al., 2008; McGrath et al., 2008), and shows a need for prioritized studies of gene function.

In contrast to Trichodesmium metatranscripts, undersampling of the mRNA pool resulted in primarily singleton or doubleton sequences in the library, with no dominant transcript pathway observed. Most gene transcripts were of unknown function or conserved hypothetical proteins. A dominant set of recognizable non-Trichodesmium transcripts were those for DNA repair and replication, suggesting that the populations of bacteria surrounding Trichodesmium colonies were actively growing. There were also a number of transcripts related to transport and signal processing (for example, tonB-dependent receptors), suggesting the dominance of these processes to the physiology of cells in this consortium or microniche.

Given the dearth of functional sequences that can be identified, it is striking that there are also a number of genes with known responsiveness to phosphorus deficiency. This sample was taken from an oligotrophic region of relatively low phosphorus and increased Trichodesmium derived nitrogen inputs, which may have increased phosphorus demand in the bloom sample. Consistent with this, there is a transcript encoding an alkaline phosphatase enzyme, which hydrolyzes phosphate from dissolved organic phosphorus, and there are several transcripts encoding exopolyphosphatases (ppX), an enzyme involved in the breakdown of polyphosphate stores. A recent proteome study in the low-phosphorus Sargasso Sea identified phosphate binding and transport, a dominant feature of the SAR11, Prochlorococcus and Synechococcus metaproteome (Sowell et al., 2009). In this study, there are similarly a number of transcripts associated with the high-affinity uptake of phosphate, including pstsS/sphX and a possible pstB transcript in Crocosphaera. These genes are typically encoded in a co-transcribed cluster (pstSCAB) that is upregulated to cope with low phosphorus availability. The Crocosphaera pstB gene is in a cluster with pstS in the WH8501 genome, the later of which has been shown to be upregulated by P deficiency in culture studies (Dyhrman and Haley 2006). Taken together, the data suggest that this consortium express phosphorus-regulated transcripts to meet phosphorus demand through the breakdown of phosphorus stores, the hydrolysis of dissolved organic phosphorus, and the high-affinity uptake of phosphate.

In oligotrophic systems, arsenate concentrations in the ocean can be similar to that of phosphate, and it has widely been hypothesized that under these conditions, arsenate would be transported into marine microbial cells through high-affinity phosphate transport systems, thus necessitating the induction of arsenate detoxification strategies (Figure 5). The speciation of arsenate in these systems suggests microbial transformation and variability in these transformations coincident with Trichodesmium blooms (Cutter and Cutter, 2006), but the genes for these pathways have not been examined in any detail in field populations. In addition to the transcripts for high-affinity phosphate transport described above, the metatranscriptome contains transcripts for arsenate reductase, and the arsA and arsB genes, which encode an arsenite efflux pump. These arsenate-related genes are typically induced by phosphorus deficiency to cope with the transport of arsenate through the pstSCAB-encoded high-affinity phosphate transport system. Although all of the transcripts in this pathway (Figure 5) are not encoded in the same taxa, their presence highlights the potential coupling of phosphorus stress and arsenate detoxification in the upper water column of oligotrophic systems and in association with Trichodesmium communities. The presence of ars and pst genes, like all genes in this study, among transcripts may also be influenced by the relative taxonomic composition of Trichodesmium colonies.

Figure 5
figure 5

A schematic of a putative phosphate/arsenate uptake and detoxification pathway. Transcripts for the highlighted proteins, including the high-affinity binding (PstS) and uptake (PstB) of phosphate, the reduction of arsenate (ArsC) and an ArsAB arsenite efflux pump are present in the prokaryotic sequences from the Trichodesmium bloom samples.

Full size image

In contrast to prokaryotic genes, reads matching microbial eukaryotic genomes were well annotated, with only 26% of day and 16% of night sequences matching genes of unknown function. Photosynthesis-related genes among eukaryotic microbial genes represented a much larger portion of the day sample (20 of 199) than the night (one of 128). Other informative genes like cytochrome c oxidase 1 (cox1) and cytochrome b (cob) were also abundant in the day (18 of 199) and night (15 of 128) samples. cox1 and cob are the mitochondrial genes that have been shown useful in DNA barcoding of fungi (Seifert et al., 2007), microalgae (Robba et al., 2006) and potentially unicellular algae (Evans et al., 2007; Lin et al., 2009).

Most mRNAs matching sequences in nr at E-values >0.001 but bit scores ⩾40 were proteins of unknown function (398 of 602 day and 513 of 719 night). Several genes had a large number of matches, comprising up to 22 % of all mRNAs in this category (Table 2). Cytoplasmic antigen 1 was the most expressed annotated protein in the nr database libraries, representing about 10% of the category in both day and night. The large transcript abundance of some highly represented organisms in this comparison, such as Branchiostoma floridae (Table 2), likely does not represent their actual abundance in the sample. It is possible that as the Trichodesmium bloom was raft-like (Figure 2a), it could provide shelter to B. floridae eggs or larvae. However, it is more likely that sequences most closely related to B. floridae are an artifact of disproportionately distributed eukaryotic genes in the non-redundant database (John et al., 2009). Model organisms studied in molecular genetics (for example, Xenopus laevis, Mus musculus and Drosophila sp.) and organism with their whole genomes sequenced (for example, B. floridae; Putnam et al., 2008) represent a large amount of Genbank's sequence coverage. Therefore, many of the sequences we annotated could represent different organisms where genes have not yet been characterized.

Unannotated sequence reads

The large number of putative mRNA sequence reads with no strong matches to microbial proteins suggests that the majority of putative mRNA in Trichodesmium colonies is made up from microorganisms for which genome data are not available, are transcripts from non-encoding intergenic regions (including small RNAs; Shi et al., 2009), or that the length of the sequence reads (175–177 bp average, shorter than the average for all sequence reads) was too short to assign reads to proteins at our E-value cutoff. Our results are in line with recent reports of putative small RNAs that indicated that they can comprise of a substantial proportion of transcripts from marine plankton (Gilbert et al., 2008; Shi et al., 2009). We compared sequences not matching the cultivated prokaryotic, eukaryotic or viral proteins with the GOS assembled protein and nucleotide databases (Venter et al., 2004; Rusch et al., 2007) at CAMERA to determine whether they were present in other open ocean microbial populations. Of 4417-day and 4476-night reads not matching microbial, metazoan or viral proteins, 2083-day and 2762-night sequences matched assembled nucleotide sequences at E <10−3 in the GOS survey. Of these, 929-day and 1144-night reads matched unannotated proteins in the assembled proteins database from GOS. In addition 2216-day and 2838-night sequences matched GOS reads, of which most were from the Sargasso Sea, followed by those from the Indian Ocean and the Galapagos Islands. However, the mean similarity of metatranscripts was not the same among different regions (Figure 6). The least similarity between metatranscripts and GOS reads was for matches to the Galapagos Island and Polynesian Archipelago sequences. However at E-values <0.001, all matches were >91% similar at the nucleotide level to the GOS sequences. This observation is surprising as these regions are more likely similar to the habitat from which the Trichodesmium samples were retrieved. These results confirm that the majority of sequences from the Trichodesmium colonies match uncultivated microorganisms (the GOS samples were from the 0.1–0.8-μm size fraction), which use proteins that share little homology to cultivated microorganisms.

Figure 6
figure 6

Number of matches to the Global Ocean Survey sequence reads at different locations (left axis) and mean similarity to reads (right axis).

Full size image

The large proportion of unrecognized mRNAs (that is, those not matching either eukaryotic microbial or prokaryotic proteins or the GOS dataset) may in part be explained by the presence of genes from metazoa inadvertently collected along with Trichodesmium colonies. Microscopic examination of Trichodesmium colonies from this station showed the presence of several metazoa (Figure 2), which is not surprising as the sample was collected at night when there are more zooplankton in surface waters. Trichodesmium is also colonized by several genera of copepods, which utilize the colonies as pseudobenthic substrates for laying eggs and for food (O’Neil and Roman, 1994; O’Neil et al., 1996). No complete marine zooplankton genomes are publicly available against which the metatranscriptomes could be compared. However, we compared the unannotated metatranscripts with the non-redundant protein database at NCBI, which showed 123-day and 178-night mRNAs having homology to metazoan proteins, including those from arthropods and echinoderms. It is possible that the remaining matches with no homology to sequenced metazoa may be proteins from organisms for which no close relatives are available, or to eukaryotic organisms with no fully sequenced genomes (for example, radiolarians or dinoflagellates).

Comparison of metatranscripts to metatranscriptomes in other studies

Despite having similar dominant transcript orthologs, the overall profile of gene expression of the Trichodesmium colonies was different from free-living pelagic communities (Frias-Lopez et al., 2008; Hewson et al., 2009; Poretsky et al., 2009b; M Vila-Costa and MA Moran, unpublished data) and coastal seawater (Gilbert et al., 2008; R Poretsky and MA Moran, unpublished data) investigated elsewhere (Figure 7). The difference between free-living and Trichodesmium transcript inventories was mostly driven by higher frequency of ribosome synthesis and oxidative phosphorylation genes, but fewer nitrogen metabolism, propanoate metabolism, porphyrin and chlorophyll metabolism, and pyruvate metabolism genes. As previous coastal seawater (Poretsky et al., 2005), eukaryotic (John et al., 2009) and soil (McGrath et al., 2008) metatranscriptomes had limited sequence depth (400, 232 and 48 sequences, respectively), direct comparison with our metatranscriptomes on the basis of gene orthology is not possible. However, the metatranscriptomes in these studies generally had dominant fractions of genes involved in central metabolism, protein synthesis, and transport and binding proteins, and were therefore distinct from the Trichodesmium colony gene expression profiles in this study.

Figure 7
figure 7

Heat map of dominant (>1%) transcript orthologs in Trichodesmium colony and free-living communities reported elsewhere. Columns in the heat map are clustered by Whittaker Index of similarity and unweighted pair-group-mean average (UPGMA) based on the frequency of transcript associated with each KEGG pathway. The total number of orthologs compared was different between samples (n=508 Trichodesmium colony day, n=417 Trichodesmium colony night, n=14 905 Stn ALOHA day (Poretsky et al., 2009a, 2009b), n=13 841 Stn ALOHA night (Poretsky et al., 2009a, 2009b), n=1725 South Pacific day (Hewson et al., 2009), n=1192 South Pacific night (Hewson et al., 2009), n=4145 Stn ALOHA night (Frias-Lopez et al., 2008), n=29 144 Fjord Mesocosm (Gilbert et al., 2008), n=45 461 Sargasso Sea day (Vila-Costa and Moran, unpublished data) and n=21 788 Sapelo Island day (Poretsky and Moran, unpublished data). For the Sargasso Sea and Sapelo Island samples, which are as yet unpublished, ‘control’ samples were used for comparison.

Full size image

Conclusions

Our data are the first on the composition of a transcriptionally active community of a pelagic microniche and suggest that Trichodesmium colonies represent microhabitats for diverse co-occurring heterotrophic bacteria, eukaryotes and phage. Although overall types of genes that are expressed as a dominant fraction of total mRNA inventories within colonies reflects those of free-living microorganisms, the distinct pattern of gene expression suggests that the habitat within colonies leads to different metabolic processes. For example, Trichodesmium transcripts related to P stress (and consequently As detoxification) response indicate that local enrichment of N leads to local P limitation. Therefore, the pervasive observation of variable and diverse populations residing on the Trichodesmium colonies with different metabolism from organisms in pelagic habitats suggests that the relevant scale of processes such as elemental cycling is intimately associated with colonies in the phycosphere or ‘trichosphere’ of the cyanobacterium. Finally, our observations show that although a large number of transcripts do not share homology with known proteins of sequenced microorganisms, they share similarity with genome fragment inventories of open ocean communities awaiting annotation of further sequencing of representative genomes.