Distribution Analysis of Hydrogenases in Surface Waters of Marine and Freshwater Environments

Background Surface waters of aquatic environments have been shown to both evolve and consume hydrogen and the ocean is estimated to be the principal natural source. In some marine habitats, H2 evolution and uptake are clearly due to biological activity, while contributions of abiotic sources must be considered in others. Until now the only known biological process involved in H2 metabolism in marine environments is nitrogen fixation. Principal Findings We analyzed marine and freshwater environments for the presence and distribution of genes of all known hydrogenases, the enzymes involved in biological hydrogen turnover. The total genomes and the available marine metagenome datasets were searched for hydrogenase sequences. Furthermore, we isolated DNA from samples from the North Atlantic, Mediterranean Sea, North Sea, Baltic Sea, and two fresh water lakes and amplified and sequenced part of the gene encoding the bidirectional NAD(P)-linked hydrogenase. In 21% of all marine heterotrophic bacterial genomes from surface waters, one or several hydrogenase genes were found, with the membrane-bound H2 uptake hydrogenase being the most widespread. A clear bias of hydrogenases to environments with terrestrial influence was found. This is exemplified by the cyanobacterial bidirectional NAD(P)-linked hydrogenase that was found in freshwater and coastal areas but not in the open ocean. Significance This study shows that hydrogenases are surprisingly abundant in marine environments. Due to its ecological distribution the primary function of the bidirectional NAD(P)-linked hydrogenase seems to be fermentative hydrogen evolution. Moreover, our data suggests that marine surface waters could be an interesting source of oxygen-resistant uptake hydrogenases. The respective genes occur in coastal as well as open ocean habitats and we presume that they are used as additional energy scavenging devices in otherwise nutrient limited environments. The membrane-bound H2-evolving hydrogenases might be useful as marker for bacteria living inside of marine snow particles.


Introduction
The composition of earth's atmosphere is the result of a number of concurring processes and a matter of continuous change. Especially the amount of trace gases governs important aspects of the gas cover of our planet, such as its retention capacity of heat or the amount of ozone present. After methane, hydrogen is the second most abundant trace gas in the atmosphere, making up around 0.5 ppm to 0.6 ppm [1,2].
Approximately 90% of hydrogen evolution is due to photochemical oxidation of hydrocarbons such as methane in the atmosphere, the combustion of fossil fuels and biomass burning. Natural evolution results from volcanic activity, the nitrogen fixation process in legumes and an uncharacterized source in the oceans. The latter comprises the majority with around 6% (6 Tg per year [3]).
The removal of hydrogen is either due to its reaction with hydroxyl radicals in the atmosphere or by its reaction with hydrogenases in the soil. In particular, hydrogen uptake into the soil is responsible for the largest term with an estimated 75% to 77% globally [1][2][3][4][5]. This is further corroborated by the lower average concentration of hydrogen found on the northern hemisphere, with its larger landmass [1]. Hydrogen uptake was attributed to aerobic hydrogen-oxidizing bacteria and extracellular enzymatic activity. Abiotic removal has been previously considered since hydrogen concentrations are below the threshold level found for cultures of aerobic hydrogen oxidizing bacteria that still maintains growth [6].
In contrast to soil, supersaturating concentrations of hydrogen have been measured in aquatic environments. In all cases, concentrations were highest at the surface and steeply decreased down to the thermocline while the deep ocean is undersaturated. Although a systematic analysis is not available it appears that surface waters of tropical and subtropical oceans are generally hydrogen sources [7][8][9]. In contrast, concentrations lower than the expected atmospheric equilibrium have been observed in higher latitudes and both hydrogen uptake and production vary depending on the season [10,11]. In some fresh water lakes supersaturation has also been found [12], with a maximum at dawn [13]. The highest hydrogen concentrations were in the upper water column, which correlated with the maximum of primary production [13,14].
Marine hydrogen uptake has been attributed to particulate fractions of 0.2 mm to 5 mm in size [11] and, like in freshwater lakes, most probably correlates with aerobic hydrogen-oxidizing bacteria [13]. Hydrogen production in the oceans was found to depend on solar radiation and clearly shows a diurnal variation with a maximum around noon [8,9]. Since the nitrogenase inevitably produces at least one molecule of hydrogen per dinitrogen reduced to ammonia, cyanobacterial nitrogen fixation is thought to be the major source of hydrogen in these oceanic regions. Studies on heterocystous cyanobacteria demonstrated that hydrogen cycling by these strains is highly effective, although under CO 2 -limitation around 0.1 nmol H 2 h 21 (mg chlorophyll) 21 escapes to the environment [15]. In contrast to this, in-situ measurements of Trichodesmium thiebautii (former Oscillatoria thiebautii), which is one of the major oceanic N-fixing strains, questioned whether its hydrogen evolution is actually sufficient to explain the concentrations found [16].
Recently it was shown that photochemical production of hydrogen from chromogenic dissolved organic matter can contribute, at least in part, to hydrogen production in fresh water lakes as well as coastal seawater [17]. Therefore, abiotic sources should be taken into account.
In the microbial world hydrogen is a valuable energy source that is exchanged efficiently between different prokaryotes and anaerobic eukaryotes. Some produce hydrogen while fermenting whereas others capture it to drive anaerobic or aerobic respiration and make use of its energy. A wealth of different enzymes called hydrogenases have been found in microorganisms that are able to split or form hydrogen [18,19].
Hydrogenases are classified according to their metal content into the Fe-, FeFe-, and NiFe-varieties. Fe-hydrogenases are confined to the methanogenic archaea and FeFe-hydrogenases occur in bacteria and anaerobic eukaryotes. NiFe-hydrogenases are separated into 4 different groups and are widespread in archaea and bacteria [19,20]. Most purified hydrogenases are only active under anoxic conditions, but there are some NiFehydrogenases from aerobic H 2 -oxidizing bacteria that are able to oxidize hydrogen at ambient oxygen concentrations [21].
Although hydrogenases have been investigated for a long time in a variety of different microorganisms it is rather difficult to deduce their physiological function on the basis of their classification alone. In Table 1 a tentative assignment of their metabolic roles is given. However, this assignment needs to be treated cautiously since several studies found surprising variations. Hydrogenase 2 of E. coli belongs to the group 1 H 2 -uptake hydrogenases and was originally described as H 2 -oxidizing enzyme [22]. In contrast, recent electrochemical data suggests that the hydrogenase 2 is working as a bidirectional enzyme [23]. Another interesting variance was found in case of the group 4 membranebound H 2 -evolving hydrogenase. In many cases these enzymes seem to be used under fermentative conditions to generate a proton gradient (e.g. [24]) but in other cases they might be used to oxidize H 2 and reduce ferredoxin with the concomitant use of a proton gradient [25] or even for H 2 uptake in N-fixing bacteria [26]. Systematic studies concerning the distribution of hydrogenases in different habitats to unravel their ecophysiological role are not yet available. Apart from the investigation of some specific soil hydrogenases [27,28] only two studies attempted the amplification of FeFe-hydrogenase sequences from microbial mats [29,30]. Although these works showed a surprising variety of these hydrogenases the short sequences amplified preclude any assignment of their function.
The hydrogen concentrations found in a variety of surface waters prompted us to investigate the presence and distribution of all known hydrogenases in marine and freshwater environments. Moreover, the ecological distribution of their genes was analyzed to collect valuable hints for their physiological functions and their oxygen tolerance.
To this end we analyzed the distribution of hydrogenases in cyanobacteria since they are one of the largest prokaryotic groups that occur in aquatic surface waters. The search was then expanded to the complete genomes of bacteria isolated from marine surface waters (http://www.ncbi.nlm.nih.gov/sutils/ genom_table.cgi [31]) and the global ocean sampling metagenomic database (http://camera.calit2.net/) [32][33][34] for all the families of hydrogenases as classified by Vignais et al. [20] and Vignais and Billoud [19]. In parallel, we investigated DNA isolated from samples taken from the North Atlantic, Mediterranean Sea, North Sea, the Baltic Sea and the fresh water lakes Westensee and Selenter See in Northern Germany for the presence of the genes of the bidirectional NAD(P)-linked hydrogenase. Our results reveal that these enzymes are surprisingly widespread in surface waters and a clear bias toward waters with terrestrial influence is obvious.

Distribution of hydrogenases in cyanobacterial genomes
Cyanobacteria are known to harbor two different NiFehydrogenases. One is called bidirectional (group 3d) since it can produce or take up hydrogen, depending on the physiological conditions and the other is an uptake hydrogenase (group 2a) that is linked to the nitrogen fixation process and seems to be confined to diazotrophic strains [35,36]. A phylogenetic analysis revealed a close ancestry of both hydrogenases to the filamentous anoxygenic photosynthetic bacteria (the former green non-sulfur bacteria) [37].
A search of genebank (http://www.ncbi.nlm.nih.gov/) and cyanobase (http://bacteria.kazusa.or.jp/cyanobase/) for all available cyanobacterial sequences revealed the presence of the bidirectional NAD(P)-linked hydrogenase (the large subunit HoxH was used in the BLAST search [38]) in all the freshwater strains and all the strains isolated from microbial mats, salt marshes, and the intertidal zone ( Table 2). In contrast, only four out of the seven available coastal genomes harbor the gene for the bidirectional enzyme and it was completely absent in oceanic strains. Genomestreamlining and iron limitation [39] in the open ocean could be used as arguments for the absence of the bidirectional hydrogenase genes in the picoplanktonic Prochlorococcus and Synechococcus strains. But even the typical open ocean strains Crocosphaera watsonii and Trichodesmium erythraeum with genome sizes above 6 Mbp do not harbor this hydrogenase, although both have the uptake hydrogenase, which has an iron requirement similar to the bidirectional enzyme ( Fig. 1 and Table 2). In addition the unicellular marine strain UCYN-A that lacks photosystem II shows an extremely reduced genome and still contains the hupgenes [40].
All the completely sequenced cyanobacterial strains that harbor the bidirectional hydrogenase genes also harbor the gene of a pyruvate:flavodoxin/ferredoxin oxidoreductase (PFOR), nifJ. In two genomes (Synechococcus WH 5701 and Arthrospira maxima), this gene is either part of the hyp-gene cluster or in close proximity to the hox-genes, suggesting that the birdirectional hydrogenase is used to dispose of electrons during fermentation via a PFOR-like enzyme ( Table 2).
The occurrence of the uptake hydrogenase (HupL, group 2a) in cyanobacteria does not correlate with a specific habitat but with the diazotrophy of the respective strains, as indicated by the presence of the nitrogenase genes (e.g. NifD)( Table 2). Of the completely sequenced genomes two Synechcococcus strains isolated from a hot spring and Cyanothece sp. PCC 7425 harbor the nitrogenase genes but no uptake hydrogenase. This confirms the previous finding of a marine nitrogen-fixing Synechococcus strain without an uptake hydrogenase [37].
Cyanothece sp. PCC 7425 is the only strain containing the genes of the bifunctional NAD(P) linked hydrogenase (group 3b)( Table 2) but expression and metabolic activity of this enzyme have not yet been demonstrated.

Distribution of hydrogenases in genomes of heterotrophic bacteria isolated from marine surface waters
Representatives of each of the hydrogenase classes were used to search the completely sequenced prokaryotic genomes in the genebank (Table 3). Of the approximately 1210 prokaryotic genomes (as of March 2010) 149 were isolated from marine surface waters and in 33 of these genomes, one or several hydrogenases occur, making up 22% of the total ( Table 4, Table  S1 supporting information). Since a number of the analyzed genomes is still not complete, this proportion is a minimum estimate. If divided into coastal and open ocean isolates, 25% of the coastal and 14% of the open ocean strains have hydrogenase genes.
The genomes of two Shewanella strains (ANA-3 and MR-4) have all the genes necessary for the expression of a FeFe-hydrogenase. Since this type of hydrogenase is extremely sensitive against and irreversibly inactivated by oxygen [41], this is a surprising finding. However, it should be noted that one strain (ANA-3) has been isolated from a wooden pier that might have been occupied by biofilms that could become anaerobic and the other strain (MR-4) was isolated from the Black Sea, which is the world largest anoxic basin [42]. Therefore, both are considered exceptions and will not be discussed any further.
The genomes of the Roseovarius group contain large gene clusters with the membrane-bound hydrogenase in conjunction with a sensor hydrogenase and the whole complement of the twocomponent system (Fig. S1, supporting information). The sensor enzyme is a receptor that enables the cells to detect hydrogen in the environment and to activate transcription of the hydrogenase structural genes [43][44][45][46]. The same gene clusters also contain a number of additional genes that encode for proteins such as HupK that have been shown to be necessary for the production of an oxygen tolerant hydrogenase in R. eutropha [47,48]. Table 2. Occurrence of the bidirectional NAD(P)-linked hydrogenase (HoxH) and the membrane-bound uptake hydrogenase (HupL) in cyanobacteria. The genomes of the Vibrionaceae harbor a membrane-bound H 2evolving hydrogenase (Fig. S2, supporting information) and a second membrane-bound hydrogenase. This is the necessary combination that can be used under anaerobic conditions to establish a proton gradient by hydrogen cycling in a single cell [49].
Additionally, the genome of N. caesariensis (former Oceanospirillum [50]) is worth mentioning. It contains a membrane-bound enzyme, a cyanobacterial like uptake hydrogenase, a sensor, and a bifunctional hydrogenase. A phylogenetic analysis confirmed that the HypX encoded in its genome belongs to the group of hydrogenase maturation factors (Fig. S3, supporting information). HypX was shown to render the soluble hydrogenase of the Knallgas bacterium Ralstonia eutropha oxygen insensitive [51]. The membrane-bound hydrogenase of N. caesariensis is a close relative of the same hydrogenase of R. eutropha (Fig. S4, supporting information), which is evidence that this bacterium and the Roseovarius strains are able to perform aerobic hydrogen oxidation in marine environments.

Distribution of hydrogenases in metagenomic databases
Single bacterial strains allow a detailed analysis of part of the genomes that occur in the specific environment they have been isolated from. However, isolated strains provide only a glimpse on the genetic diversity that might be present in the habitat from which they originate, given that most microbial strains are unculturable [52][53][54]. Therefore, we searched the global ocean sampling database (GOS) [32][33][34] with the same representative hydrogenases as given in Table 3 and the representatives of the small hydrogenase subunits ( Fig. S5 to S7, supporting information).
This database contains millions of sequence reads that have been obtained mostly from biological samples with a particle size of 0.2 to 0.8 mm. Due to this size fractionation, the major proportion of the sequences belongs to Pelagibacter ubique and the Prochlorococcus/Synechococcus group of cyanobacteria [33]. Since the large number of sequences in the Sargasso Sea metagenome belonging to the Shewanellaceae and the Burkholderiaceae was discussed to be a contamination [55] Station 11 was not included in the analysis.  We could not detect any cyanobacterial bidirectional hydrogenase in the samples taken from the open ocean. All the cyanobacterial HoxH sequences that could be found in the database are from a single sample taken at Punta Comorant, a hypersaline pond with low oxygen levels [56] on the Galapagos Islands (Fig. 2). These sequences were most similar to the available bidirectional hydrogenases of Synechococcus strains (Fig.  S8, supporting information). Thus, the GOS sampling and sequencing effort should have been able to capture any HoxH sequence present in the Prochlorococcus/Synechococcus group. Although it has to be taken into account that environmental sequencing does not capture 100% of the present DNA sequences it seems highly probable that this cyanobacterial hydrogenase is absent in these strains in these environments as already deduced from the whole genomes (Table 2, Table S1, supporting information).
These findings are also corroborated when looking at the hoxH sequences of the Burkholdericeae. Although these bacteria make up a major fraction of all the oceanic metagenome sequences, there are only representatives from Punta Cormorant with this hydrogenase (Fig. 2), whereas no sequences of this group have been retrieved from the open ocean. Altogether 48 hoxH sequences could be found but apart from three coastal stations (Mangrove on Isabella Island, Cape May and Dirty Rock), which accounted for 4 sequences all of the other 44 were exclusively from Punta Cormorant. This confirms the presence of hoxH in  shallow coastal environments and ponds in a variety of different bacterial groups. The largest group of sequences in the metagenome database were those of the membrane-bound NiFe-hydrogenases. Again most of the 51 sequences were found at Punta Cormorant, although 11 sequences were detected in the datasets of coastal stations (New Harbor, Dirty Rock, Yucatan Channel, Nags Head, a Mangrove on Isabella Island) and two were found in the open ocean (outside Seychelles and 250 miles of Panama) (Fig. 3).
Cyanobacterial-like uptake hydrogenases could also be found in the metagenomic dataset (Fig. 4). Because of the size fractionation (0.2-0.8 mm) most of the larger diazotrophic cyanobacteria have been excluded from this analysis. Therefore, although many of the samples have been taken in regions known to be inhabited by this cyanobacterial group only two sequences could be retrieved from the whole dataset. A total of 35 sequences could be found. Most of these sequences originate from coastal sites (28)  Searches for the small hydrogenase subunit genes retrieved 23 sequences of the bidirectional NAD(P) + -linked hydrogenases, 37 of the membrane bound H 2 uptake hydrogenases and 18 of the cyanobacterial-like uptake hydrogenases. In all these cases the numbers are close to the expected number when comparing the gene sizes of the respective large and small hydrogenase genes ( Fig.  S5 to S7, supporting information).
Sequences of the oxygen sensitive FeFe-hydrogenases retrieved from the GOS database were from a Mangrove (Isabella Island) and the hypersaline pond at Punta Cormorant. In all other samples no FeFe-hydrogenase was found (Fig. 5) and none of the archaebacterial hydrogenases were found in the metagenome sequences.
Recently large amounts of metatranscriptomics data became available (e.g. [57]). A search of the respective dataset revealed the presence of three transcripts of membrane-bound H 2 -uptake hydrogenases. One transcript was most similar to a cyanobacterial uptake hydrogenase, one to the Flavobacteriaceae and one to the Bradyrhizobiaceae. In this dataset only samples from the open ocean are available.
Detection of sequences of the bidirectional NAD(P)linked NiFe-hydrogenase in the North Atlantic, Mediterranean Sea, North Sea, Baltic Sea, and two freshwater lakes Although all NiFe-hydrogenases share two characteristic motifs with altogether four cysteins at the N-and C-terminus for the binding of the NiFe active site, it is impossible to design degenerated primers that bind to the genes of all different classes of these enzymes. Therefore, we limited our effort to a single class and constructed degenerated primers specific for the bidirectional NAD(P)-linked hydrogenases of cyanobacteria, the Chloroflexaceae and some proteobacteria. In cyanobacteria this enzyme is known as the bidirectional hydrogenase. It is closely related to the soluble hydrogenase of Ralstonia eutropha and the respiratory complex I [58,59].
We collected surface water from Stollergrundrinne outside the Kielfjord (Baltic Sea), in the Norderpiep west of Büsum (North Sea) and two freshwater lakes in northern Germany, Westensee and Selenter See. These samples were sequentially filtered on 10 mm and 0.2 mm filters and DNA isolated from the retained material. In samples from all these locations we could detect hoxH. In Fig. 6 the distribution of sequences on the different bacterial groups is shown for the different stations.
From the Baltic Sea as well as the fresh water lakes we could amplify a large number of cyanobacterial hoxH that are most similar to the Chroococcales (most closely related to Cyanothece, Microcystis and  Table 4. Cont. Synechocystis) or the filamentous, heterocystous Nostocaceae. In the North Sea the a-proteobacterial group Rhodobacteraceae made up the same proportion as all the cyanobacterial sequences taken together. From the freshwater mesotrophic lakes Westensee and Selenter See we could only amplify cyanobacterial hoxH (Chlorococcales, Nostocaceae and Oscillatoriales) and in each case some sequences of methylotrophic bacteria and Dictyoglomaceae.
In contrast to this, all attempts to amplify sequences of the bidirectional NAD(P)-linked hydrogenases from the samples taken in the North Atlantic off the west African coast and the Ionian Sea (Mediterranean Sea) were negative. This corroborates that the open ocean and marine oligotrophic waters are devoid of this hydrogenase type.

Discussion
Any conclusion concerning the activity of a gene from its environmental distribution is hampered by the fact that it is not necessarily expressed in a specific environment. Genomes might have genes in store that are not necessary to survive under the present-day conditions, but can be used to invade other niches or to prepare the organism for a drastic change. In the case of the   (Table 3) was used for the search and a total of 51 sequences has been retrieved. On the right the number of sequences from the different sampling stations is shown. doi:10.1371/journal.pone.0013846.g003 distribution of hydrogenases found in this work, this scenario seems highly unlikely. For several reasons described in detail below, we think that biological hydrogen production and consumption, as depicted in Fig. 6, might be common in a large number of marine and freshwater habitats.
All strains from the open ocean were free of the bidirectional NAD(P) linked hydrogenase. Neither the cyanobacterial genomes nor all of the heterotrophic bacteria (Table 2 and Table S1, supporting information) or the metagenomic sequences harbor this hydrogenase. In addition, our efforts to amplify these hydrogenase genes from the North Atlantic or the Mediterranean Sea were unsuccessful. Since the diazotrophic cyanobacterial strains and the heterotrophic bacteria from the open ocean have other types of hydrogenases, there is no selection pressure against these enzymes per se. However, there is a clear bias of the bidirectional type to environments such as coastal marine waters, ponds, freshwater lakes and microbial mats ( Table 2, Fig. 2, Fig. 6 and Table S1, supporting information), where cyanobacteria and heterotrophic bacteria might encounter micro-oxic or anaerobic conditions. In cyanobacteria this type of enzyme was shown to be activated under anaerobiosis and to be responsible for fermentative hydrogen production [60]. This is corroborated by the distribution of the PFOR gene, nifJ, in the same cyanobacteria ( Table 2).
Starting from anaerobiosis, the bidirectional hydrogenase is known to be used as an electron valve, when cells switch from fermentation to photosynthesis [61][62][63][64]. These findings might explain the high hydrogen concentration found in the morning hours in a eutrophic lake that coincided with the phytoplankton maximum [13]. Oxygen depletion due to high respiratory activity during the night could have activated the hydrogenase in this zone and elicited a fermentative hydrogen production in the dark that continued at dawn until the next morning when photosynthesis resumed, thus causing supersaturating H 2 concentrations. A   (Table 3) was used for the search and a total of 10 sequences have been found. On the right the number of sequences from the different sampling stations is shown. doi:10.1371/journal.pone.0013846.g005 similar diel variation of hydrogen concentrations has also been described for cyanobacterial mats (see e.g. [65]).
In both cases, hydrogen production is certainly not confined to the resident cyanobacteria but can also result from the activity of algae and other heterotrophic bacteria living in the same community.
The large number of genomes of marine bacteria from surface seawaters containing the membrane-bound H 2 -uptake hydrogenase is remarkable. A search of the current marine metatranscriptomics data [57] revealed the expression of these hydrogenases in cyanobacteria as well as other bacteria in the open ocean.
The membrane-bound hydrogenase gene clusters found in the Rhodobacteraceae (Fig. S1 supporting information) include all the accessory genes that are known from the membrane-bound hydrogenase of R. eutropha. One of the four hydrogenases of N. caesariensis and the hydrogenases of the Roseovarius strains are closely related to this hydrogenase as revealed by phylogenetic analysis (Fig. S1 and S4, supporting information). This type of enzyme is known to be oxygen insensitive and was shown to be active at ambient oxygen concentrations [66,67]. Electrochemical investigations of this hydrogenase found measurable hydrogen uptake down to levels of 1 to 10 nM [67], which is well in the range of H 2 concentrations in surface waters. One of these strains (Roseovarius sp. HTCC 2601) was isolated from the Sargasso Sea, but all of the others were from coastal areas. In these regions, this a-proteobacterial subclass makes up as much as 24% of the bacterioplankton [68] and therefore, their hydrogenases might be widespread in these environments.
Mycobacteria, known to colonize aquatic ecosystems, take up hydrogen in the same concentration range under aerobic conditions [69], supporting the notion that hydrogen consumption in these environments is a common microbial feature. Even though the supersaturating concentrations found in surface waters are below the threshold necessary to support growth exclusively on H 2 , hydrogen uptake could add to the ability to survive in a variety of these habitats. Similar suggestions have already been made for hydrogen uptake for long-term survival of bacteria [70] and for the ability to oxidize carbon monoxide in the coastal ocean [71,72]. These suggestions coincide with the aerobic hydrogen uptake demonstrated for particle sizes between 0.2 and 5 mm in coastal waters [11]. This trait is especially important for litho-and heterotrophic bacteria that have to capitalize on as much of the available energy supply as possible, but can be disregarded by photoautotrophs like cyanobacteria.
Bacterial activity was found to be capable of depleting oxygen in marine organic aggregates. In particles as small as 1.5 mm, anoxic conditions emerged. In the same aggregates no methanogenic or sulfate-reducing bacteria could be detected [73]. Our results suggest that these anaerobic microniches might be specifically occupied by bacteria of the Vibrionaceae (Fig. S2, supporting information). Since their membrane-bound H 2 -evolving NiFehydrogenases are encoded in conjunction with subunits of the formate dehydrogenase it seems highly likely that it performs the formate:hydrogen lyase reaction. This reaction is well known from E. coli, where it detoxifies formate produced during fermentation, evolves hydrogen and might be involved in an additional energygenerating step [24]. The membrane-bound hydrogen uptake hydrogenase encoded in the same genomes would allow hydrogen cycling and might be used for additional net transport of protons across the cell membrane [49].
The Altermonadaceae are widespread in marine waters. Two different ecotypes have been sequenced, one is predominant in surface waters whereas the other is known from the deep Mediterranean Sea. The deep ecotype was originally found to harbor the genes of the membrane-bound H 2 uptake hydrogenase but our analysis and that of others [74] also found the same sequences at the surface of the Sargasso Sea. It was speculated that these two strains are separated by either being associated with small aggregates (surface type) or large aggregates (deep ecotype) [75]. This might be further support for the use of hydrogenases in transiently anoxic microniches in the ocean.
The diel variation of the H 2 concentration in marine surface waters [8,9] that parallels solar radiation is still awaiting conclusive explanation. Nitrogen fixation is a major source of hydrogen in terrestrial ecosystems [4]. In-situ measurements of the diazotrophic cyanobacterium T. thiebautii suggest that it is a negligible source of hydrogen in the Sargasso Sea [15]. Therefore, nitrogen fixation by filamentous cyanobacteria is an insignificant source of H 2 in aquatic ecosystems. Interestingly, a unicellular marine diazotrophic cyanobacterium has been shown to be devoid of the uptake hydrogenase [37] and to produce hydrogen while fixing nitrogen [76]. In general unicellular cyanobacteria perform a temporal separation of the oxygen sensitive energy consuming nitrogen fixation process and oxygenic energy generating photosynthesis between night and day, but some strains also fix nitrogen during the light phase [76,77]. Unicellular strains are known to provide a considerable part of fixed nitrogen in marine waters [78,79] and might therefore be responsible for part of the evolved H 2 . The newly discovered unicellular cyanobacteria without photosystem II [40,80] harbor the genes of the cyanobacterial uptake hydrogenase (Table 2), which is most similar to those of the Cyanothece group (Fig. S4, supporting information) as expected. Therefore, these strains should be able to recycle the H 2 evolved by the nitrogenase.
The distribution of cyanobacterial nitrogen fixers in the ocean and their seasonal abundance are poorly characterized although qPCR data has shown that all groups are widely distributed [81,82]. One investigation suggests that their distribution is patchy and their rate of nitrogen fixation highly variable [79] and might therefore result in hydrogen evolution in some parts and very low or no evolution in other parts.
Although unicellular nitrogen-fixing cyanobacteria might be responsible for hydrogen evolution in some regions, part of the H 2 produced during the day might be of photochemical origin, such as dissociation of organic matter by UV light [17].
Coastal waters are rich in hydrogenase sequences, as suggested by our analysis of complete genomes ( Table 2, 4, Table S1, supporting information), and the number of sequences we could amplify of a single class of NiFe-hydrogenases from the North Sea and the Baltic Sea (Fig. 5). The apparent scarcity of sequences from coastal samples in the GOS database can be explained by the filtration procedure. Since mainly particle sizes between 0.2 and 0.8 mm have been used for DNA isolation many of the coastal bacteria and particle associated bacteria have been excluded from the analysis. We hypothezise that the membrane-bound H 2 evolving hydrogenase in the genomes of the Vibrionaceae might be used as indicator for bacteria that colonize the inner parts of organic aggregates and thus, have not been sequenced yet in the GOS database.
Our analysis shows that the genetic repertoire of bacteria from surface waters of different environments enables them to produce hydrogen either by their nitrogenase, by hydrogenases linked to fermentative pathways (such as the bidirectional NAD(P) linked hydrogenase), or the membrane-bound H 2 -evolving hydrogenase. A number of bacteria could oxidize hydrogen as an energy source probably down to the lower nM range and might be responsible for biological hydrogen consumption in freshwater and marine systems.
This study intends to deliver a first key to the elucidation of the underlying biological processes of hydrogen turnover in aquatic ecosystems. Whether a specific body of water is a hydrogen sink or source will depend on a number of factors such as primary production, nitrogen fixation, the concentration of photodegradable organic compounds and organic particles, and the availability of electron acceptors. This is the first evidence that microorganisms can be an integral part of hydrogen turnover in marine waters, but much more remains to be learned. This is especially true when considering oxygen minimum zones [83] that have not been investigated for the presence of hydrogen or hydrogenases until now.

Sample collection
Samples were collected from the surface. In the North Sea water was collected in the Norderpiep (54u139N/8u279E), in the Baltic Sea it was collected in the Stollergundrinne (54u299N/10u139E) and from the freshwater lakes Selenter See (54u18925N/ 10u28953E) and Westensee (54u17953N/9u57909E) at least four times a year from every season. These samples were sequentially filtered on 10 mm and 0.2 mm filters with a peristaltic pump 620 S (Watson-Marlow Bredel).
Samples from the Mediterranean Sea were taken from the Ionian Sea at station 2 (36u419N/21u399E), station 3 (36u509N/ 21u319E), station 5.2 (36u379N/21u179E) and station 6 (36u429N/ 21u049E). In this case 5 l water from a depth of 5 m was filtered on 5 mm and then on 0.2 mm.
The samples from the North Atlantic were taken during the Poseidon 284 cruise at 18uN/30uW, 25uN/30uW and 29u/30uW in April 2002.
For DNA isolation the UltraClean TM Soil DNA Kit (Mo Bio, Carlsbad CA, USA) was used.

DNA amplification and sequence analysis
Sequences of the bidirectonal NAD(P)-linked hydrogenase were amplified with the primers HoxH-f GTATYTGYGGYATT-TGTCCTGT and HoxH-r GGCATTTGTCCTRCTGYATG-TGT were used. Prior to 40 cycles of the program the DNA was denatured for 5 min at 95uC. The temperature program was as follows: 30 sec at 95uC, 40 sec at 50uC, 2 min at 72uC. In a final step the temperature was kept at 72uC for 10 min. The reaction contained 0.5 mM of the two primers, 0.2 mM of dNTPs, 2.5 mM MgCl 2 , 0.025 U/ml Taq polymerase (MBI Fermentas, St. Leon-Roth, Germany) and 10x buffer as recommended by the manufacturer in a total volume of 50 ml. Of each sample different amounts of DNA between 2 and 100 ng were tested as template. If no PCR product was detected DNA concentrations were increased at least 10 times. Positive controls were run in parallel to prove the efficiency of the PCR. The approximate size of the product is around 1190 bp and covers close to 84% of the hoxH gene.
The resulting PCR products were ligated into the pCRII-topo (Invitrogen), sequenced with the Big-Dye Kit, and applied on a 96 capillary sequencer (3730 DNA Analyzer, Applied Biosystems).
If possible contigs were assembled from the obtained sequence data and the respective sequences deposited in the genebank (Accession numbers GQ454414 to GQ454443 and GU238237 to GU238258) including two additional cyanobacterial hoxH sequences of Aphanothece halophytica and Mastigocladus laminosus SAG 4.84 (Accession numbers GQ454444 and GQ454445).

Database searches
The genebank, cyanobase, and the GOS database were searched for hydrogenase specific sequences by using the hydrogenase sequences given in Table 3. Retrieved sequences were either run against the genebank by using the BLAST algorithm [38] to deduce the closest homolog or searched for the signature sequences as given by Vignais and Billoud [19] to unambiguously classify the respective hydrogenase. In case of the GOS database, the sequences found were aligned, and, if possible, larger contigs were formed from the same sampling station and used for all further analysis.

Phylogenetic analysis
In the case of critical candidates or unclear phylogenetic affiliation phylogenetic trees were used. Sequence alignments were made with ClustalW [84]. After manual optimization and removal of gaps from the alignments, parsimony, maximum likelihood, and distances were calculated with the 3.63 release of the PHYLIP package [85], using the Jones-Taylor-Thornton matrix and the algorithm of Fitch and Margoliash [86]. Maximum parsimony and distances were calculated for 1000 bootstraps and maximum likelihood for 100 bootstraps. The Unix-cluster at the computer center of the University of Kiel was used for most of the calculations. The resulting trees are given in Fig. S3 to S5 (supporting information).  Figure S1 Structure of the gene cluster of the membrane bound hydrogen uptake NiFe-hydrogenase of marine Rhodobacteraceae and the delta-proteobacterium Neptuniibacter caesariensis. The structural genes of the hydrogenase (hupS, hupL and hupZ the membrane bound cytochtrome) are shown in blue. Red genes (hoxAJBC) are involved in the regulation of the hydrogenase. HoxJ encodes a histidine kinase that is known to interact with a hydrogen sensor encoded by hoxB and hoxC and regulates the activity of the response regulator encoded by hoxA. HupK might encode a protein necessary to express an oxygen-tolerant hydrogenase. Accessory genes known to be necessary for this type of membrane hydrogenase are shown in grey, whereas grey patterned genes are general accessory genes for all NiFe-hydrogenases. Genes depicted in green are putative proteases that cleave the C-terminus of the hydrogenase. HypX of Ralstonia eutropha is known to render its soluble hydrogenase oxygen tolerant. Found at: doi:10.1371/journal.pone.0013846.s002 (0.06 MB DOC) Figure S2 Structure of three hydrogenase gene clusters of Vibrioanceae isolated from marine environments that are most similar to the energy converting H2-evolving NiFe-hydrogenases. The color code is the same as in Figure S1. Genes shown in plaid are part of the fromate dehydrogenase. FhlA is the transcriptional activator of the formate-hydrogen lyase. Those in black and greyblue are additional subunits of the whole complex.  Figure S5 Distribution of small subunits of the bidirectional NAD(P)+ linked hydrogenase found in the GOS database of the different prokaryotic groups. The small subunit gene, hoxY, of Synechocystis has been used for the search. All genes have been retrieved form Punta Comorant, a hypersaline pond on the Galapagos Islands. Found at: doi:10.1371/journal.pone.0013846.s006 (0.06 MB DOC) Figure S6 Distribution of small subunits of the membrane bound H2 uptake hydrogenasses found in the GOS database of the different prokaryotic groups. The hupS sequence of Desulfovibrio vulgaris was used for the search. On the right the number of sequences from the different sampling stations is shown. Found at: doi:10.1371/journal.pone.0013846.s007 (0.08 MB DOC) Figure S7 Distribution of small subunits of the cyanobacteriallike uptake hydrogenase found in the GOS database of the different prokaryotic groups. The small subunit gene, hupS, of Nostoc sp. PCC 7120 has been used for the search.