Giant Hydrogen Sulfide Plume in the Oxygen Minimum Zone off Peru Supports Chemolithoautotrophy

In Eastern Boundary Upwelling Systems nutrient-rich waters are transported to the ocean surface, fuelling high photoautotrophic primary production. Subsequent heterotrophic decomposition of the produced biomass increases the oxygen-depletion at intermediate water depths, which can result in the formation of oxygen minimum zones (OMZ). OMZs can sporadically accumulate hydrogen sulfide (H2S), which is toxic to most multicellular organisms and has been implicated in massive fish kills. During a cruise to the OMZ off Peru in January 2009 we found a sulfidic plume in continental shelf waters, covering an area >5500 km2, which contained ∼2.2×104 tons of H2S. This was the first time that H2S was measured in the Peruvian OMZ and with ∼440 km3 the largest plume ever reported for oceanic waters. We assessed the phylogenetic and functional diversity of the inhabiting microbial community by high-throughput sequencing of DNA and RNA, while its metabolic activity was determined with rate measurements of carbon fixation and nitrogen transformation processes. The waters were dominated by several distinct γ-, δ- and ε-proteobacterial taxa associated with either sulfur oxidation or sulfate reduction. Our results suggest that these chemolithoautotrophic bacteria utilized several oxidants (oxygen, nitrate, nitrite, nitric oxide and nitrous oxide) to detoxify the sulfidic waters well below the oxic surface. The chemolithoautotrophic activity at our sampling site led to high rates of dark carbon fixation. Assuming that these chemolithoautotrophic rates were maintained throughout the sulfidic waters, they could be representing as much as ∼30% of the photoautotrophic carbon fixation. Postulated changes such as eutrophication and global warming, which lead to an expansion and intensification of OMZs, might also increase the frequency of sulfidic waters. We suggest that the chemolithoautotrophically fixed carbon may be involved in a negative feedback loop that could fuel further sulfate reduction and potentially stabilize the sulfidic OMZ waters.


Introduction
Eastern Boundary Upwelling Systems are found along the westward shelves of the continents in both the Atlantic and the Pacific Ocean. They are characterized by high photoautotrophic primary production, which is driven by the upwelling of nutrientrich waters [1]. The produced biomass supports large fish populations in these regions, underlining the importance of Eastern Boundary Upwelling Systems in providing a source of food for mankind [2][3][4][5][6]. However, a significant proportion of the produced biomass also sinks through the water column and is remineralized in subsurface waters, contributing to the oxygen (O 2 ) depletion in intermediate water depths of these regions. Combined with large-scale ocean circulation patterns and poor ventilation of intermediate waters with the ocean surface, the remineralization of rich organic matter leads to oxygen-depleted zones of varying intensities [7][8][9][10]. These oxygen-depleted waters, also referred to as oxygen minimum zones (OMZs), are found in the eastern tropical North and South Pacific, and to a lesser extent in the eastern tropical North and South Atlantic [11]. In addition to OMZs found in regions of strong upwelling, oxygen-depleted waters are also present in the northern Indian Ocean and in enclosed basins like the Baltic and the Black Sea [12].
The OMZ off Peru, Chile and Ecuador in the South Pacific Ocean is the largest oceanic area where O 2 concentrations are reported to fall below the detection limit of the most sensitive O 2 sensors (,10-100 nM) [13][14][15][16]. In the absence of O 2 , organic carbon degradation has been historically attributed to heterotrophic denitrification, the reduction of nitrate (NO 3 2 ) to dinitrogen gas (N 2 ) [17][18][19]. Some in situ experiments have shown active heterotrophic denitrification in OMZ waters [20,21], however numerous studies have demonstrated that anammox, the anaerobic oxidation of ammonium (NH 4 + ) with nitrite (NO 2 2 ) to N 2 , is responsible for the major loss of fixed nitrogen from the OMZ off Namibia [22], Oman [15], Peru [23][24][25] and Chile [26]. As anammox is an autotrophic process, its dominance over heterotrophic denitrification questions our understanding of organic matter remineralization in OMZ waters. This is supported by the hypothesis that oxygen-depletion leads to a shift of the microbial community from organoheterotrophs to chemolithotrophs [27,28].
In anoxic sediments, the main process for the degradation of organic carbon is considered to be microbial sulfate (SO 4 22 ) reduction to elemental sulfur (S 0 ) and hydrogen sulfide (H 2 S) [29]. The release of large quantities of the toxic H 2 S from these underlying sediments can lead to the occasional build-up of high concentrations of H 2 S in bottom waters [30][31][32][33]. On the other hand, it has also been suggested that H 2 S build-up in oceanic waters could be caused by SO 4 22 reduction of pelagic microorganisms within the water column [34,35]. A recent study suggested that an active, but cryptic sulfur cycle is present in non-sulfidic subsurface waters in the eastern tropical South Pacific OMZ off northern Chile [14]. According to this hypothesis, SO 4 22 reduction and consequently H 2 S formation would take place in the water column, even when thermodynamically more favourable electron acceptors like NO 3 2 or NO 2 2 are still present. However, the resulting H 2 S would be rapidly re-oxidized to S 0 or SO 4 22 , such that the two processes are in steady-state, thereby preventing an accumulation of H 2 S [31,36].
The presence of chemolithoautotrophic c-proteobacteria involved in sulfur cycling (e.g. related to the uncultured SUP05 cluster bacterium, referred to henceforth as SUP05) in non-sulfidic OMZ waters supports the hypothesis of a cryptic sulfur cycle [8,14,[37][38][39][40]. Studies conducted during occurrences of sulfidic events in the Benguela Current upwelling OMZ and in a seasonally anoxic fjord in Canada demonstrated that c-proteobacteria related to SUP05 are highly abundant in sulfidic waters and that they probably detoxified the waters via the chemolithotrophic oxidation of H 2 S coupled to the reduction of NO 3 2 [31,41], a pathway termed sulfur-driven autotrophic denitrification. In addition, chemolithoautotrophic aand especially eproteobacteria detected in the sulfidic waters in the Benguela Current upwelling OMZ as well as in the sulfidic zones of the Baltic and the Black Sea could also be important members of microbial communities in sulfidic plumes [31,42,43].
The initiation, termination and frequency of sulfidic events in oceanic OMZs are so far poorly understood, and H 2 S in the water column has been mostly observed in enclosed basins like the Baltic Sea [43][44][45], the Black Sea [46][47][48], the Cariaco trench off Venezuela [49,50] and the Saanich Inlet in Canada [41,51]. With the exception of a few studies mentioning the characteristic odor of H 2 S and anecdotal reports of Peruvian fishermen on 'black' fishing gear in relation to the so-called 'aquajes' conditions for the OMZ off Peru, sulfidic waters have not been measured in the Pacific Ocean so far [31,34,35,52].
Negative consequences on fish stocks and quality of life along the populated coastal upwelling regions are potentially severe, because H 2 S is highly toxic to animals and humans and has already been invoked as the cause for occasional but massive fish kills in African shelf waters [53][54][55]. The anticipated decrease in O 2 concentrations and the increase in water column stratification, as predicted from global change [11], as well as local eutrophication [27,35], might lead to more frequent and intense depletion of O 2 and of alternate electron acceptors (e.g. NO 3 2 and NO 2 2 ), favouring the development of sulfidic waters within OMZs [31]. Given that the detoxification of sulfidic water is a microbial process, it is important to assess the phylogenetic structure and the metabolic response of the endemic microbial community to the accumulation of H 2 S.
We explored the microbial community structure and its transcriptional activity in such a sulfidic event off the coast of Peru with high-throughput metagenomic and metatranscriptomic sequencing. We present rate measurements of carbon dioxide (CO 2 ) fixation and nitrogen transformation processes as well as total bacterial cell counts with flow-cytometry. Several of the proteobacterial taxa that were dominant in the sulfidic waters were expressing genes involved in the sulfur cycle, which reflected various metabolic strategies for H 2 S oxidation. Our data further suggests that the microbial communities were responsible for considerable light-independent CO 2 fixation.

Results and Discussion
Description of the sampling site During RV Meteor cruise M77/3 on the Peruvian shelf in January 2009 we found sulfidic waters stretching from Lima to Paracas National Reserve southwest of Pisco ( Figure 1). O 2 concentration in shelf waters in the study area were generally below the detection limit of our microsensor (0.5-1 mM) at water depth below 20 m (from 12u S to 14u S; Figure 1A), while NO x (the sum of NO 3 2 and NO 2 2 ) was heavily depleted in the water column throughout the transect from 20-60 m to the bottom (from 12u309S to 13u509S; Figure 1B), mirror-imaging the distribution of H 2 S ( Figure 1C). Sulfidic waters were first detected on January 9 th south of Lima and seemed to have persisted until the end of the cruise, when H 2 S-containing waters covered ,5500 km 2 of the shelf. The thickness of the sulfidic layer was about 80 m ( Figure 1C), yielding the largest sulfidic plume (,440 km 3 ) ever reported for oceanic waters. We calculated a H 2 S content of ,2.2610 4 tons. The total area affected by H 2 S may have been even larger, as we did not map the extent of the sulfidic plume into the protected area of the Paracas National Reserve.
Remote satellite sensing revealed large patches (50-150 km 2 ) of turquoise discoloured surface-waters, attributable to the formation of colloidal S 0 upon H 2 S oxidation [56,57] off Lima as well as in Paracas National Reserve during our sampling campaign ( Figure  S1). The larger extension of H 2 S in deeper waters when compared to the colloidal S 0 in the surface indicated that most of the H 2 S was oxidized in subsurface waters, similar to the observations from the Benguela upwelling system [31]. Colloidal S 0 plumes measuring up to 500 km 2 were observed in the same region again in May 2009, indicating that the occurrence of S 0 plumes detected in January 2009 was not a unique event ( Figure 1E). This suggests that sulfidic waters in the OMZ off Peru might be more frequent and persistent than originally thought.
A vertical profile of the sulfidic water column (station 19), sampled during the upcast with a pump-CTD on January 9 th , 2009 at a site located ,15 km offshore Lima (12u21.889S, 77u0.009W, ,100 m water depth; Figure 1D) was the target of a detailed analysis. The surface mixed layer was shallow with the thermocline at about 10 m water depth ( Figure 2D). The surface temperature (.16uC) was only ,2uC warmer than the bottom waters and the salinity (34.95-34.97) changed merely slightly with depths, which indicated an active upwelling of subsurface waters. Even the surface waters were characterized by low O 2 conditions down to 40 mM (or ,15% saturation; Figure 2A). During the downcast, O 2 decreased at the thermocline and dropped to about the detection limit of our amperometric O 2 microsensor (0.5-1 mM) at around 20 m. Nevertheless, trace amounts of O 2 (,1 mM) were still detected with some variability down to ,40 m water depth. These low concentrations of O 2 were close to detection limit of our sensor and we cannot rule out whether this was due to water advection caused by the CTD rosette or to a memory effect of the sensor. However, using a highly-sensitive selfcalibrating Switchable Trace amount Oxygen (STOX) sensor with a detection limit of ,50 nM [13,16] during the upcast of the CTD rosette, O 2 was undetectable below 20 m, and therefore we defined this zone as anoxic. We detected large vertical movement in the oxycline from 5-18 m due to internal waves, which may Phosphate (PO 4 32 ) and NH 4 + concentrations remained high (both around 3 mM) and stable throughout the water column, with NH 4 + only having a minor drop in concentrations near 50 m ( Figure 2C). NO 3 2 concentrations (detection limit ,0.1 mM) were lower than those measured in the southern part of the study area and lower than expected from strong upwelling regions in general. Highest NO 3 2 concentrations with ,5 mM were found in surface waters, but dropped rapidly below1 mM, just beneath the oxycline at ,19 m ( Figure 2B). Detectable concentrations of NO 3 2 (ranging from 0.1-0.2 mM) were measured down to 59 m with the exception of 52-55 m, where a small increase up to 0.7 mM was observed. NO 2 2 (detection limit ,0.01 mM) was high (1.5-3 mM) in surface waters down to 26 m. Trace concentrations (,25 nM) were measurable down to 50 m and again from 67-81 m. Nitrous oxide (N 2 O) concentrations ranged between 20-40 nM from 15-80 m and dropped below the detection limit closer to the bottom of the water column ( Figure 2B). H 2 S was first detected (with both microsensor and wet chemistry) at 26 m and increased steadily, reaching a concentration of 4.2 mM at 48 m ( Figure 2A). This maximum was followed by a rapid drop in concentrations to below 0.1 mM at 52-53 m, before increasing again to about 2.6 mM at 95 m, approximately 5 m above the sediment. H 2 S concentrations directly at the sediment-water interface were probably even greater, but were not measured in this study.

Phylogenetic diversity of the microbial community
Based on the monitoring of O 2 and H 2 S during the upcast, we defined three zones within the water column, where we carried out a detailed sampling: the oxic surface (5 m sample, where sampling was stopped when internal waves decreased O 2 concentrations to below 30 mM), the upper boundary of the anoxic zone (15 and 20 m samples) and the sulfidic zone (30,40,50,60,80 and 100 m samples). Both a hierarchical clustering approach with a statistical analysis of taxonomic assignments and a non-parametric Multidimensional Scaling indicated that the selected sample groups were justified (an ANOSIM test using a Bray-Curtis distance measure showed a Global R value of 0.83 and a significance level of 0.1%, Figure S2).
Using a 98% similarity cut off, the metagenomes and metatranscriptomes accounted for an average of 263,606 (DNA) and 98,785 (RNA) unique sequences (clusters) per depths (Table  S1). A total of 4809 (DNA) and 3872 (RNA) different taxa were identified using BLAST-searches, revealing a highly diverse microbial community at all depths (Table S2). The taxonomic composition based on all rRNA genes (5,923 sequences) using BLASTn-searches against the SILVA database is shown in Figure 3A and based on the metagenomes (1,882,842 sequences, excluding all rRNA genes) and the metatranscriptome (421,528 sequences, excluding all rRNAs) using BLASTx-searches against the non-redundant database of NCBI in Figure 3B and 3C. A large percentage of the sequences found in both the metagenomes and metatranscriptomes had no significant match against the nonredundant database of NCBI. On average, 49% of the sequences remained unidentified, which is comparable to other studies that utilized high-throughput sequencing technologies in marine habitats [37,40,[58][59][60].
The community structure presented a stable and uniform distribution at the phylum-level, especially within the sulfidic zone ( Figure 3). The metagenome data suggested that the microbial community was overall dominated by proteobacteria (16.6-34.1% of all DNA sequences, including all unidentified sequences); while the Bacteroidetes/Chlorobi-group was the second largest group we could identify (3.9-7.4%). In oxic and anoxic waters, both aand c-proteobacterial sequences were abundant (6.9-16.4%), similar to previous findings from the OMZs off northern Chile [38] and the Arabian Sea [61]. In sulfidic waters, c-proteobacteria were clearly dominating (up to 23.2%) and we further found a significant increase in the frequencies of dand e-proteobacterial sequences (1.9-4.0%), which were much less abundant in the 5 and 20 m samples. The dand e-proteobacterial sequences were even more abundant within the rRNA gene dataset ( Figure 3A) when compared to the metagenomes ( Figure 3B), probably due to the lack of representatives genomes for these groups.
The metatranscriptomes showed a more variable picture of the microbial community ( Figure 3C). In surface waters eukaryotic sequences formed the largest identifiable group (5%), while at all other depths c-proteobacteria were dominating (13.3-30.5%). In sulfidic waters e-proteobacterial transcripts were further identified in relatively high, but also more variable proportions (5.0-13.2%), when compared to the metagenomes. Notably, 'other bacteria', summarizing 19 bacterial phyla, were present at all depth, but never exceeded 5.8% of the sequences, in both the metagenome and the metatranscriptome datasets.
A more detailed analysis of the metagenomes revealed that the oxic surface waters harboured several different photoheterotrophic organisms ( Figure S3). Prokaryotes similar to Candidatus Pelagibacter sp. HTCC7211 accounted for 1.6% and relatives of Candidatus Pelagibacter sp. HTCC1002 made up 0.4% of all DNA sequences, which is in agreement with other studies conducted in OMZs [14,37,40,61]. Additionally, photosynthetic Synechococcus spp., which are also known to be present in OMZ waters [62] accounted for about 0.4% of the sequences. However, the most abundant single taxon identified in the oxic surface metagenome had high similarity to the uncultured SUP05 cluster bacterium (1.9%), a chemolithoautotrophic sulfur oxidizer, which has been detected previously in oxygen-depleted [14,37,38,40] and sulfidic waters [31,41,42,63]. In the metagenome sample from the anoxic zone (20 m), SUP05 was also the dominant taxon with 6.6% of all DNA sequences ( Figure 4A and S3). At 20 m and below, cproteobacterial sulfur oxidizers (GSO) related to gill symbionts of deep-sea hydrothermal-vent clams, Candidatus Ruthia magnifica str. Cm (2.6%) and Candidatus Vesicomyosocius okutanii HA (1.4%) [64,65] became increasingly abundant. This indicated that the GSO-community at our sampling site was composed of at least three separate taxa. In the sulfidic zone, the dominance of the GSO-group was even higher, reaching a maximum of 17% of all DNA sequences at 50 m (Figure 4 and S3). Other common microorganisms in the metagenome were similar to the eproteobacterium Sulfurovum sp. NBC37-1 (up to 1.7%) and to the d-proteobacterium Desulfobacterium autotrophicum HRM2 (up to 1.4%) [66][67][68]. In all sulfidic depths, organisms related to SUP05, R. magnifica, V. okutanii, Sulfurovum and D. autotrophicum were the five most abundant organisms that we identified with BLASTsearches.
In contrast to the metagenomes, the metatranscriptomes reflect the suite of genes that were expressed in the microbial community at the time of sampling and therefore displayed a more variable picture of the microbial community, showing to some extent its actual metabolic activity. In the oxic surface waters, we found RNA sequences similar to the archaeal ammonia-oxidizer Nitrosopumilus maritimus SCM1 and the a-proteobacterium Magnetospirillum gryphiswaldense to comprise the largest single taxa (0.7 and 0.4%). N. maritimus is considered a 'classical' inhabitant of the oxycline in OMZ waters [14,37,39,40]. At 20 m, RNA sequences similar to the c-proteobacteria Marinomonas sp. MWYL1 and Neptuniibacter caesariensis were most abundant (1.3 and 1.2%).
Similar to the metagenome assignments, the GSO-group was predominant in the metatranscriptomes throughout the sulfidic zone. However, the composition of the GSO-group was variable and changed with depths. While SUP05 was virtually undetectable using BLAST-searches in the 5, 20 and 40 m metatranscriptomes (0-0.1%; Figure 4B and S3), GSO-related RNA sequences at these depths could be almost exclusively assigned to the relatives of R. magnifica and V. okutanii. In all other metatranscriptomes from sulfidic waters (50, 60 and 80 m), sequences similar to SUP05 represented the most abundant identifiable taxon (5.5-11%).
Also similar to the metagenomes, the five most abundant organisms detected in the metatranscriptomes within sulfidic waters (except at 40 m) were related to SUP05, R. magnifica, V. okutanii, Sulfurovum and D. autotrophicum. We also identified several other taxa in the metatranscriptomes that were poorly represented in the metagenomes, some of which were closely related to members of hydrothermal vent communities. At 40 m, organisms related to the e-proteobacterium Arcobacter butzleri RM4018 accounted for 2% of all RNA sequences and at 50 m the cproteobacterium Colwellia psychrerythraea 34H accounted for 3%. In deeper parts of the water column (60 and 80 m), organisms similar to the e-proteobacteria Sulfurimonas gotlandica GD1 (1.1-1.3%) and Arcobacter nitrofigilis DSM 7299 (0.9-1%) were detected. Although sequences similar to the anammox-planctomycete Candidatus Kuenenia stuttgartiensis have been found in high abundances in OMZ waters [14,37,40], they were relatively rare within our samples, never exceeding 0.7% of all DNA and 0.8% of all RNA sequences.

Metabolic activity and functional diversity of the microbial community
General activity patterns. To assess the functional diversity of the microbial community in detail, we used three different approaches to investigate our sequence data. The BLASTsearches were supplemented by scans of our sequences with profile hidden Markov models of the ModEnzA Enzyme Commission (EC) groups [69] and of the Pfam protein families [70]. Furthermore, we recruited the DNA and RNA sequences onto the genomes of the five organisms most often recognized by our BLAST-searches, SUP05, R. magnifica, V. okutanii, Sulfurovum and D. autotrophicum [41,[64][65][66][67]. For these genome recruitments, we calculated the expression-ratio, a measure of the enrichment of selected transcripts over the corresponding genes, normalized to the total pool of all protein-coding sequences ( Figure 5 and S4). In general, the genome recruitment plots showed high expressionratios for ribosomal proteins, DNA and RNA polymerases, cell division proteins and transcription and translation factors indicating a growing microbial community (data not shown). Similarly, the collection of all abundant EC numbers also suggested an overall active microbial community ( Figure S5). Sequences encoding for ubiquitous proteins related to general metabolic activity like DNA and RNA polymerases, DNA topoisomerases and adenosinetriphosphatases (a general ATP-binding and hydrolyzing motif in sequences which lack further specific functional information) were among the most abundant at all depths.
Hydrogen sulfide sources. H 2 S formation through microbial SO 4 22 reduction commonly occurs in anoxic marine sediments, where it is considered to be the main heterotrophic process for the degradation of organic carbon [29]. The sedimentary flux has been shown to be the main source of H 2 S in the water column during sulfidic events in the Benguela Current upwelling system [30,31]. However, at times water column SO 4 22 reduction could also contribute significantly to the H 2 S accumulation in oceanic waters [34,35], as demonstrated in the ,2000 m thick anoxic water column in the Black Sea, where pelagic SO 4 22 reduction rates ranged between 0.01-3.5 nmol l 21 d 21 [71].  Moreover, H 2 S formation from SO 4 22 has been measured even in the presence of more favourable electron acceptors (NO x ) in the OMZ waters off northern Chile after preincubations with H 2 S [14].
We investigated the sulfur cycling by identifying genes and transcripts indicative of specific metabolic functions related to sulfur transformation processes in our collection of sequences and used flux calculations to estimate the sedimentary source of H 2 S. D. autotrophicum, one of the most abundant organism we identified in our sequence dataset, is a metabolically versatile SO 4 22 reducing marine d-proteobacterium, which can completely oxidize organic carbon compounds to CO 2 , but is also capable of growing autotrophically on hydrogen (H 2 ) [67]. The genome recruitment plots of D. autotrophicum ( Figure S4B) show regions of the genome mostly related to energy metabolism and nutrient cycling. High expression-ratios for key sulfur metabolizing enzymes like the dissimilatory sulfite reductase (dsrABD), the adenylylsulfate reductase (aprAB) and the sulfate adenylyltransferase (sat2) suggest that D. autotrophicum could have been reducing SO 4 22 and thus may have contributed to the formation of H 2 S at our sampling site. However, since many sulfur cycling proteins (e.g. dsr and apr) can function in both the oxidation and reduction of sulfur species [72,73], other chemolithoautotrophic metabolic processes (e.g. the disproportionation of sulfur compounds resulting in the simultaneous formation of H 2 S and SO 4 22 ) could have been catalyzed by these enzymes at the time of sampling [74]. The presence of large plumes of colloidal S 0 ( Figure S1) in the study area would have supported this chemolithoautotrophic reaction.
A numerical compilation of all sequences affiliated to known SO 4 22 reducers and sulfur oxidizers is shown in Figure 6 [14,72,73]. Although the abundance of SO 4 22 reducing organisms throughout sulfidic waters was not correlated with the H 2 S concentrations, it is likely that they contributed to SO 4 22 reduction and H 2 S formation in the water column at the time of sampling. However, assuming the maximum SO 4 22 reduction rates (1.3-12 nmol l 21 d 21 ) measured in OMZ waters off northern Chile [14], it would take more than one year (in the absence of any oxidation) to accumulate the H 2 S concentrations reported in our study. In contrast, SO 4 22 reduction in sediments underlying the eastern tropical South Pacific OMZ is generally very high (10-30 mmol m 22 d 21 ) [75] and thus the most likely source of the H 2 S we measured in the water column.
Assuming steady state conditions, we used the density structure and the H 2 S concentration gradient (Figure 2A) in the bottom water to estimate a turbulent diffusion of ,10 24 m 2 s 21 and, subsequently, a sedimentary efflux of ,2 mmol H 2 S m 22 d 21 . This is well within the estimates based on sedimentary flux calculations and SO 4 22 reduction rate measurements (1-11 mmol m 22 d 21 ) from sediments directly underlying sulfidic events [76]. Moreover, the repeated observations of the H 2 S maxima in bottom waters along the transect ( Figure 1C) point towards the sediment as the main H 2 S source.
We observed a second distinct H 2 S maximum in the water column at ,48 m with no direct contact to the sediment. However, the salinity and the corresponding PO 4 32 and NH 4 + concentrations ( Figure 2C and 2E) indicated that the upper H 2 S layer was most probably created by lateral advection of nearby bottom waters that had recently been in contact with sulfidic sediments rather than by production of H 2 S within the water column. Sulfur oxidation. The largest functional group of microorganisms detected in sulfidic waters were c-proteobacterial sulfur oxidizers. Figure 5 shows the expression-ratio for selected genome regions of the three most abundant GSO-representatives. The recruitment of the sequences onto separate genomes supported the presence of at least three distinct taxa within the GSO-community (similar to SUP05, R. magnifica and V. okutanii) at our study site, which were actively growing and metabolizing. Figure S4A further depicts a genome recruitment plot for e-proteobacterial Sulfurovum. The recruitment of sequences onto the genome of Sulfurovum was much less extensive than for the GSO-group, and showed low coverage especially in the oxic and anoxic depths (5 and 20 m).
A large number of transcripts were recruited onto genes (if present in the genomes) of the reverse dissimilatory sulfite reduction (dsr -oxidation of intracellular S 0 ) and the periplasmic sulfur oxidation (sox -S 2 O 3 22 oxidation) pathways, both encoding for enzymes involved in the oxidation of reduced sulfur compounds. Additionally, we found high expression-ratios for the sulfide:quinone oxidoreductase (sqr -H 2 S oxidation to S 0 ), the adenylylsulfate reductase (apr) and the sulfate adenylyltransferase (sat -latter both SO 3 22 oxidation to SO 4 22 ). The abundance and coverage of transcripts matching these clusters for the GSO-group and Sulfurovum increased generally with depth, delivering the highest expression-ratios of genes encoding for sulfur oxidizing proteins within sulfidic waters. For Sulfurovum, which is thought to be capable of both sulfur oxidation and SO 4 22 reduction [66,77,78], the transcript coverage for its soxand sqr-genes suggested that it was likely acting as a sulfur oxidizer at the time of sampling.
R. magnifica, V. okutanii and Sulfurovum genomes harbour genes for different cytochrome c oxidases (cox) which are used in oxic respiration and are absent in the currently annotated version of the SUP05 genome [41,[64][65][66]. Of the transcripts recruiting to the R. magnifica genome, the cytochrome c oxidase was among the most abundant in the oxic surface. In anoxic and sulfidic waters, however, expression patterns changed and transcripts for a cbb 3type cytochrome c oxidase dominated instead. Transcripts for the cbb 3 -type cytochrome c oxidase recruiting to the genomes of V. okutanii and Sulfurovum were also among the most highly expressed in sulfidic waters.
The cbb 3 -type cytochrome c oxidase is thought to be involved in a specialized microaerobic respiration. This enzyme has an extremely high affinity to O 2 , with a K M value as low as 7 nM, allowing certain proteobacteria to colonize oxygen-limited or even presumed anoxic environments [79,80], well below the detection limit of the microsensor (0.5-1 mM) and the STOX sensor (,50 nM) we used in this study. Furthermore, it was shown that Escherichia coli cultures actively grew and respired O 2 even below the detection limit of a highly sensitive STOX sensor (#3 nM), probably using a high-affinity cytochrome bd oxidase [81].
A clear separation of the cytochrome c oxidase expression was visible between the oxic and the anoxic/sulfidic zone, directly reflecting the availability of O 2 ( Figure 6C). In oxic surface waters, the low-affinity cytochrome c oxidase was dominant and assigned mainly to eukaryotic sequences, as well as to diverse bacterial groups belonging to aand c-proteobacteria ( Figure 6D). In contrast, in sulfidic waters where the high-affinity cbb 3 -type cytochrome c oxidase prevailed, as much as 80% of the transcripts could be assigned to either cand e-proteobacteria. For the R. magnifica-like organism, which possesses both types of cytochrome oxidases, the switch in the expression from the low-affinity type in oxic surface waters (5 m) to the high-affinity type in anoxic and sulfidic waters (20-80 m) can be observed in the genome recruitment plots ( Figure 5B).
Despite the presence of H 2 S, the oxygen microsensor showed trace amounts of O 2 (,1 mM) down to 40 m water depth during the downcast, which might be an artefact from water advection caused by the CTD rosette (see section 'description of the sampling site'). Assuming that the measured O 2 concentrations were not an artifact, we calculated a vertical down flux of 0.07 mmol O 2 m 22 d 21 between 17 and 27 m. Although this O 2 flux could only account for the oxidation of ,6% of the sedimentary H 2 S flux of 2 mmol m 22 d 21 , it could partly sustain microaerobic activity in the sulfidic waters as suggested by the presence of the cbb 3 -type cytochrome c oxidase transcripts. Lateral advection of oxic waters and water exchange induced by internal waves may also have supported microaerobic activity at O 2 concentrations that would remain below the detection limit of our STOX sensor (,50 nM). However, mixing due to internal waves would have influenced only the upper part of the water column and the widespread anoxic conditions below 30 m gave little indication of lateral advection of oxic waters from 30-100 m, where the cbb 3 -type cytochrome c oxidase was nevertheless strongly expressed. Either, the O 2 was consumed shortly before our sampling campaign or advective transport of oxic waters supplied O 2 in concentrations below the detection limit of our STOX sensor. An alternative explanation for the high cbb 3 -type cytochrome c oxidase expression in the sulfidic waters could also be the use of nitric oxide (NO) as an alternate substrate instead of O 2 , as hypothesized for a cproteobacterial species [82] and supported by structural similarities of cbb 3 -type cytochrome c oxidases to bacterial nitric oxide reductases [83]. Although we did not measure NO in our study, a high expression-ratio of the genes encoding for nitric oxide reductases (norBC) were found in the genome recruitments for SUP05 (these genes are absent from currently available versions of the R. magnifica and V. okutanii genomes) in sulfidic depths, and at 60 m also for Sulfurovum. Furthermore, transcripts for the dissimilatory nitrite reductase (nirK) were highly expressed in SUP05-like organisms throughout the water column. Since this enzyme reduces NO 2 2 to NO, minor concentrations of NO might have been available for respiration processes linked to sulfur oxidation.
Alternatively to O 2 (and NO) utilization, SUP05, V. okutanii and Sulfurovum and several other e-proteobacteria [84,85] are also capable of coupling the oxidation of H 2 S to the reduction of NO x . Low but persistent concentrations of NO x were measurable at most depths down to 60 m, and nitrate as well as nitrite reductase (nar-, nap-and nir-genes) transcripts were expressed in all anoxic and sulfidic depths ( Figure 5 and S4). The genome recruitment plots for SUP05 (the genome contains genes from all three clusters) show that napand especially nir-genes were expressed, while relatives of V. okutanii and Sulfurovum showed high expression-ratios mostly for narand nap-genes, respectively. In contrast, most of the genes mentioned above are absent from the currently available version of the R. magnifica genome and it is thought that R. magnifica reduces NO x for assimilatory reasons only [65].
We could not calculate fluxes from the trace amounts of NO x below 30 m. However, at the upper boundary of the second H 2 S peak (25-29 m)   Although our results suggest that H 2 S was oxidized by both microaerobic activity and sulfur-driven autotrophic denitrification well below the oxic zone (.1 mM O 2 ), our flux calculations indicate that the larger part of the H 2 S was probably oxidized anaerobically with NO x .
Nitrogen cycling. To shed light on the nitrogen cycling carried out by the microbial community, we measured potential rates of various nitrogen transformation processes using 15 NO 3 2 , 15 NO 2 2 , 15 N 2 O or 15 NH 4 + incubations and compared them with corresponding abundances of functional genes and transcripts (the sum of BLAST-hits, EC number-and Pfam-assignments) involved in their turnover (Figure 7). Reduction of NO 3 2 to NO 2 2 was active throughout the anoxic and sulfidic zones, with the highest rates measured at 40 m (2500 nmol N l 21 d 21 ; Figure 7B). In comparison, the much lower rate measured at 30 m (150 nmol N l 21 d 21 ) might have been caused by limitations in reductants, as suggested by the much lower H 2 S concentrations at that depth. The transcript abundance for respiratory nitrate reductase (EC 1.7.99.4) peaked in the anoxic zone at 20 m (0.35% of all proteincoding sequences) and then dropped within the sulfidic waters following the decrease in NO 3 2 concentrations. Gene abundances however, increased with depth (,0.5%) and it is possible that the addition of NO 3 2 necessary for rate measurements stimulated a fast response of the microbial community yielding the actual potential for NO 3 2 reduction to NO 2 2 rather than the in situ rate. The majority of the genes and transcripts encoding for respiratory nitrate reductase at 5 and 20 m were similar to K. stuttgartiensis, while transcripts in sulfidic waters belonged to diverse groups of a-, b-, c-, dand e-proteobacteria, indicating a clear taxonomic separation with depth.
Measured rates for the subsequent steps in denitrification, the reduction of NO 2 2 to N 2 were highest close to bottom waters at 80 m (900 nmol N l 21 d 21 ; Figure 7C), whereas the transcript abundance for NO-forming cd-cytochrome nitrite reductase (EC 1.7.2.1) was highest in the oxic and anoxic zones at 5 and 20 m (0.4% and 0.2%), mirroring the availability of NO 2 2 . The majority of the transcripts at 5 and 20 m were similar to the gene from N. maritimus, while those recovered from sulfidic depths were affiliated with diverse groups of proteobacteria. We also found high expression of ammonia monooxygenase (EC 1.14.99.39) transcripts related to N. maritimus (1.5%, data not shown) in the oxic surface (5 m), which may be partly responsible for the high NO 2 2 concentrations in the surface waters. Some of the NO 2 2 reduction (N 2 production) could be attributed to anammox activity, with the highest rates measured at 30 m (250 nmol N l 21 d 21 based on 15 NO 2 2 addition, Figure 7F  ; an incubation with added 15 NH 4 did not stimulate any N 2 production, most likely due to limitations in NO x . The abundance of genes encoding for hydrazine oxidoreductase (EC 1.7.99.8) was generally very low at all depths (less than 0.015%; Figure 7F) while transcripts, mostly annotated as similar to K. stuttgartiensis, peaked in abundance at 20 and at 50 m (0.15 and 0.25%, respectively). The 50 m transcript  maximum of hydrazine oxidoreductase (we did not measure rates at this depth) is in good agreement with a small peak in NO 3 2 concentrations (to 0.7 mM) and a minor dip in NH 4 + concentrations ( Figure 2C). The transcript maximum might also be influenced by the much lower H 2 S concentrations at this depth, as anammox activity was shown to be inhibited by H 2 S [86]. N 2 O is an intermediate in denitrification (NO x to N 2 ) and we measured the reduction of N 2 O to N 2 , which turned out to be of smaller magnitude (30 nmol N l 21 d 21 ) than the NO 2 2 reduction to N 2 ( Figure 7D). The N 2 O concentrations, which ranged between 20 and 40 nM from surface to ,80 m, dropped below detection limit at 80 m ( Figure 2B), indicating either a complete reduction of N 2 O as previously observed for this area [87] or a lack of production due to the limitation in NO x . Gene and transcript abundance for nitrous oxide reductase (EC 1.7.2.4) were highest in anoxic waters (20 m) reaching 0.3% and comparable in magnitude to the abundances of sequences encoding for nitrate and nitrite reductases from the same depths.
We also conducted rate measurements of complete denitrification (NO 3 2 to N 2 ; Figure 7A). The highest rate (490 nmol N l 21 d 21 ) was found at 40 m within the first H 2 S maximum. Much lower rates were observed for the other depths (26-41 nmol N l 21 d 21 ), which might be attributed to incomplete denitrification (e.g. NO 3 2 reduction to NO 2 2 , NO or N 2 O) and potentially also to limitations in concentrations of the reductant (H 2 S).
Rates for DNRA were the lowest of the nitrogen transformations processes we measured, not exceeding 40 nmol N l 21 d 21 ( Figure 7E). Gene abundance for cytochrome c nitrite reductase (EC 1.7.2.2) was also lower than genes implicated in the other nitrogen transformation processes. Except for 50 m depth, the transcripts for this gene were even rarer than the gene abundances (less than 0.01%). This suggests only a minor role of DNRA in the microbial community metabolism at the time of sampling.
Although rate measurements do not provide information on the phylogenetic affiliation of organisms carrying out the nitrogen transformations, the predominance of sulfur oxidizing proteobacteria found in our sequence data suggests that the sulfur-driven autotrophic denitrification was most likely the dominant pathway for N-loss during our sampling campaign.
Carbon assimilation. Eastern Boundary Upwelling Systems are characterized by high primary productivity due to photosynthetic (photoautotrophic) growth in surface waters, which in turn stimulates heterotrophic respiration processes in underlying waters. However, autotrophic lifestyles have also been found in these underlying waters, e.g. by organisms responsible for nitrification [28,88], anammox [22][23][24]26] and sulfur-driven autotrophic denitrification [31,44,84]. To investigate the magnitude of inorganic carbon assimilation of the microbial community, we conducted rate measurements at selected depths with 13 Cbicarbonate incubations and compared them to the relative abundance of transcripts encoding for key carbon-fixing enzymes (Figure 8). High abundances of transcripts for ribulose-bisphosphate carboxylase/oxygenase (RuBisCo, EC 4.1.1.39) were found at all depths, but especially in sulfidic waters. At 40, 60 and 80 m transcripts encoding for RuBisCo were the most abundant to be identified in the metatranscriptomes (1.1, 2.1 and 1.6% of all protein-coding sequences, respectively; Figure 8B and Figure S5). The phylogenetic diversity of the RuBisCo transcripts varied throughout the water column. Transcripts from photosynthetic organisms such as algae and diatoms (e.g. similar to Heterosigma akashiwo, Odontella sinensis and Thalassiosira spp.) as well as cyanobacteria dominated the 5 and 20 m samples. In the sulfidic zone (40-80 m), band especially c-proteobacterial transcripts were most abundant ( Figure 8C). Approximately 25% of all RuBisCo-transcripts were most similar to a bacterial artificial chromosome-clone of unknown bacterium 560, which also appears to belong to a SUP05 genome [89]. Altogether, transcripts from c-proteobacteria contributed about 70% of all RuBisCo transcripts in sulfidic waters.
While the high proportions of RuBisCo transcripts were indicative of an active Calvin-Benson-Bassham cycle, other CO 2 fixation pathways were also represented by the presence of transcripts for key enzymes of the Arnon-Buchanan cycle (ATP citrate lyase, EC 2.3.3.8) and the Wood-Ljungdahl pathway (COdehydrogenase, EC 1.2.99.2). Transcripts encoding for these two enzymes accounted for up to 0.3% of all protein-coding sequences at 40 m depth ( Figure 8B) and were also recruited onto the Sulfurovum and D. autotrophicum genomes, respectively ( Figure S4). CO 2 fixation rates measured at selected depths with 13 Cbicarbonate incubations ( Figure 8A) were highest in the nutrientrich oxic surface waters (23 mmol C l 21 d 21 ), reflecting the dominance of large-sized eukaryotic phytoplankton. However, dark incubations with 13 C-bicarbonate yielded chemolithoautotrophic CO 2 fixation rates ranging from 0.9 to 1.4 mmol C l 21 d 21 at depths of 30, 80 and 100 m. These measured rates are comparable to chemolithoautotrophic activity found in the redoxclines and the sulfidic zones of the Baltic and the Black Sea [42,43,[90][91][92][93]. In comparison, carbon assimilation rates of 0.3 mmol C l 21 d 21 by heterotrophic bacteria, after the addition of 13 C-glucose (data not shown), was only about one third when compared with the chemolithoautotrophic CO 2 fixation. Based on total microbial cell counts, each cell fixed ,0.3 fmol C d 21 , a magnitude that was also observed in the sulfidic zone of the Baltic Sea [92].
Integrating the CO 2 fixation rates over the predicted predominantly photic zone (0-20 m) we estimated ,288 mmol C m 22 d 21 being fixed. This is in good agreement with the highest primary production rates modelled for the study area during the same period of time (250 mmol C m 22 d 21 ) [25]. On the other hand, the integrated light-independent CO 2 fixation rates in the predicted predominantly aphotic zone (20-100 m) reached ,96 mmol C m 22 d 21 . Consequently, ,25% of the total CO 2 fixation at our study site was carried out by chemolithoautotrophic microorganisms. This is similar to the dark CO 2 fixation in the redoxcline of the Baltic Sea, which was shown to contribute ,30% of the surface productivity and to the dark CO 2 fixation rates as reported for Chilean waters, which accounted for an average of ,20% of total CO 2 fixation [88,94]. The chemolithoautotrophic CO 2 fixation at our study site could represent as much as 33-53% of the estimated average CO 2 fixation rate per square meter for the Humboldt Current System (182-290 mmol C m 22 d 21 ) [2,4]. Assuming the measured rates were maintained throughout the entire sulfidic plume (,5500 km 2 ), the CO 2 fixed through chemolithoautotrophy would have been ,6.3610 3 tons C per day. In comparison, in this area an estimate of the total primary production by remote sensing is ,5.5610 5 tons C per day [2]. Consequently, the chemolithoautotrophic activity during this sulfidic event would have contributed ,1.2% of total CO 2 fixed in the Humboldt Current System off the Peruvian coast at the given time. This is intriguing, since the Humboldt Current System is one of the most productive marine systems world-wide and supports the production of more fish per unit area than anywhere else in the world [2][3][4]95]. Moreover, since the sulfidic plume may have been considerably larger than the extent we recorded and was either recurrent or prevailed for several months, the chemolithoautotrophic growth is significant in terms of carbon retention. Considering that only 15-30% of the photosynthetic surface production is actually exported to OMZ waters [25] the chemolithoautotrophic growth may act as an important, but up to now neglected factor promoting SO 4 22 reduction and thus stabilizing sulfidic conditions in OMZ waters.

Conclusions
The Eastern Boundary Upwelling System off the Peruvian coast is one of the world's most productive oceanic regions and comprises one of the largest OMZs world-wide [2,4,11]. We reported here the detection of a sulfidic plume within continental shelf waters of the Peruvian OMZ in January 2009. The sulfidic plume covered an area .5500 km 2 and contained ,2.2610 4 tons of toxic H 2 S, representing with ,440 km 3 the largest sulfidic plume ever observed in oceanic waters and the first time that H 2 S was measured in Peruvian OMZ waters.
The microbial community was largely dominated by several distinct cand e-proteobacteria related to SUP05, R. magnifica, V. okutanii and Sulfurovum, which transcribed a broad range of genes involved in sulfur (H 2 S, S 2 O 3 22 , S 0 and SO 3 22 ) oxidation. Our data suggested that these sulfur oxidizing proteobacteria probably utilized several different oxidants ranging from O 2 , NO 3 2 , NO 2 2 , NO to N 2 O to oxidize the H 2 S well below the oxic surface. While sequences related to SUP05 indicated that genes involved in the reduction of NO x and NO were being expressed, R. magnifica-, V. okutaniiand Sulfurovum-like transcripts related to a microaerophilic cbb 3 -type cytochrome c oxidase also pointed to the use of O 2 for H 2 S oxidation.
High-throughput sequencing data further showed a high abundance and transcriptional activity of d-proteobacterial SO 4 22 reducing D. autotrophicum. Transcripts recruiting to the SO 4 22 reduction genes of D. autotrophicum were found suggesting that SO 4 22 reduction may have occurred in the water column. However, our flux calculation indicated that the main source for H 2 S in the water column was sedimentary SO 4 22 reduction. Given the fact that many sulfur cycling enzymes can both oxidize and reduce sulfur species [72,73] and the presence of colloidal S 0 plumes within the sampling area as observed with remote satellite sensing, the disproportionation of sulfur compounds could also be a way of energy acquisition for some of the proposed SO 4 22 reducers [74].
High CO 2 fixation rates were measured in photic surface waters (5.8-23 mmol C l 21 d 21 ), but also in the dark sulfidic zone, ranging from 0.9 to 1.4 mmol C l 21 d 21 . Many identified microorganisms, both sulfur oxidizers and SO 4 22 reducers were expressing transcripts encoding for key carbon-fixing enzymes. The light-independent, chemolithoautotrophic CO 2 fixation is similar to observations made in permanently stratified systems like the Baltic and the Black Sea [42,43,[90][91][92][93]. If the rates prevailed throughout the entire sulfidic zone, they would have represented as much as ,30% of the photoautotrophic CO 2 fixation at that site and would have been of similar magnitude as the photosynthetic surface production that is exported to OMZ waters [25].
The presence of colloidal S 0 plumes in the study area again in May 2009 indicated that sulfidic waters in the OMZ off Peru might be more frequent and persistent than originally thought. Although the frequency and duration of H 2 S accumulations cannot be estimated from direct observations, their occurrence might increase in future resulting from eutrophication and global warming [27,31,35]. In addition, the carbon retention due to the chemolithoautotrophic activity presented in this study may enhance SO 4 22 reduction and consequently H 2 S formation and thus could act as an important mechanism to stabilize sulfidic conditions in OMZ waters, potentially reducing the liveable habitat of many higher organisms.

Sample Collection
All waters samples were collected during RV Meteor cruise M77/3 (December 27 th , 2008 to January 24 th , 2009) on the Peruvian continental shelf. Station 19 (January 9 th , 2009; 12u21.889S, 77u0.009W), located approximately 15 km off the coast of Lima was selected for detailed analysis. During the upcast, water was pumped from depth directly on board using a pumpconductivity-temperature-depth (pump-CTD) water sampler. We monitored density as well as O 2 and H 2 S to account for internal waves and moved the pump-CTD accordingly.
Samples for nucleic acid extraction were filled (oxygen-free) in 4.5 l polycarbonate bottles. For each sample, 1.5-2 l of the water was prefiltered through 10 mm pore size filters (Whatman Nuclepore Track-Etch) and then collected upon 0.22 mm pore size filters (Millipore Durapore Membrane) using a vacuum pump. Less than 18 minutes elapsed from the time when the water was pumped on board until the filters were put in microcentrifuge reaction tubes and flash frozen in liquid nitrogen. Samples for incubation experiments and nutrient analysis were taken in parallel (see description below).

RNA extraction and cDNA synthesis
DNA and RNA were extracted using the DNA/RNA-Allprep kit (Qiagen) with minor modifications in the protocol for the lyses step: The frozen filters were crushed using a disposable pestle and incubated with 200 ml lysozyme (10 mg/ml) and 1 mM EDTA at ambient temperatures for 5 minutes. Subsequently, 40 ml of Proteinase K (10 mg/ml) were added, followed by another incubation of 5 minutes at ambient temperatures. After adding 500 ml buffer RLT-Plus (containing 10 ml/ml b-mercaptoethanol) the manufacturer's instructions were followed. DNA was eluted in 150 ml elution buffer; RNA in 50 ml nuclease-free water, followed by a subsequent step of DNA digestion (Turbo DNA-free kit, Ambion). Prokaryotic rRNA was removed with mRNA only prokaryotic mRNA isolation kit (Epicentre). Further depletion of bacterial rRNA was achieved by using the Ambion MicrobExpress kit (see also section 'sequencing statistics'). Cleaned and rRNAdepleted mRNA was subjected to an in vitro transcription (amplification) step using Ambion MessageAmp. Finally, cDNA was synthesised using the Invitrogen superscript III cDNA synthesis kit with random hexameric primers (Qiagen). Throughout the procedure all DNA and RNA samples were subsequently quantified with a NanoDrop spectrophotometer and checked for degradation with a BioRad Experion Automated Electrophoresis System. Leftover reactants and reagents were removed using the PCR Mini Elute Kit (Qiagen). DNA and RNA was stored at 280uC until pyrosequencing. Throughout the whole procedure nuclease-free plastic consumables and nuclease-free water and reagents were used to hinder any possible degradation of DNA or RNA.

Sequencing
For both DNA and RNA (cDNA), 50 ml were sequenced with the GS-FLX (Roche) pyrosequencer at the Institute of Clinical Molecular Biology in Kiel. Each sample was loaded on one quarter of a PicoTiter plate (except the 5 m RNA sample, which was loaded on two quarters of a PicoTiter plate). This resulted in 1,888,768 (DNA) and 1,560,959 (RNA) sequences with an average length of 392 base pairs, accounting for 757,439,211 and 599,103,110 base pairs of sequence information, respectively (Table S1).

Sequence annotation pipeline
The sequence data was organized and analyzed with the Meta2Pro annotation pipeline [96]. All raw sequences were clustered using Cd-hit [97] with a sequence identity threshold of 98% and a word length of 8, delivering about 1,581,637 (DNA) and 592,711 (RNA) cluster representative sequences in total. The rRNA genes and rRNAs in these cluster representatives were identified by a BLASTn-search [98] against the SILVA database [99] including both prokaryotic and eukaryotic small and large ribosomal subunit sequences with a bit score cut off of 86. The bit score cut off of 86 as described earlier [100] was validated using a simulation exercise. All sequences deposited in the SILVA database (477,749 sequences) were randomly fragmented into one million sequences to simulate a pyrosequencing run. The generated fragments had a normal length distribution with mean and standard deviation values similar to those from our own dataset (mean: 420 base pairs and standard deviation: 150 base pairs). This simulated dataset was Cd-hit-clustered (as described above) and BLAST-searches against the SILVA database itself were carried out. All queries which hit themselves in the database (e.g. the sequence of origin of the query) or hit a sequence belonging to the same taxonomic lineage (allowing a mismatch of up to two taxonomic levels) were considered true positives, while those fragments that hit sequences from other taxonomic lineages were false positives. The bit score distributions for the true and false positives were binned and a threshold sweep was used to calculate the sensitivity and specificity value at each bit score threshold. A bit score value of 86 in the resulting plot of the sensitivity and specificity distributions for this threshold sweep gave a specificity of 99.35% and a sensitivity of 99.85% respectively. Hence, this cut off was used for all further analysis of the sequences against the SILVA database and also in MEGAN to make taxonomic assignments (using a minimum support of 5 and a 10% score range for its LCA algorithm) [101]. The cluster representative sequences without a hit in the SILVA database were compared against the non-redundant database from NCBI using BLASTx with a bit score cut off of 35. The top hit of each BLASTx-search was used for the functional assignment of the cluster representatives.
The cluster representative sequences without a hit in the SILVA database were further scanned with profile hidden Markov models of the ModEnzA EC groups [69] and the Pfam protein families [70]. The Pfam-hits were converted to EC numbers and along with the ModEnzA EC-hits mapped to the Kyoto Encyclopedia of Genes and Genomes (KEGG) reference pathways using the FROMP pathway mapping and visualization tool [96]. For ease of data analysis all cluster representative sequences, including clustering information from Cd-hit (cluster identification number, cluster size and all clustered nucleotide sequences), results from BLAST-, Pfam-and EC-searches as well as the taxonomic assignment from MEGAN were added to a MySQL database [96].

Sequencing statistics
Combining all DNA sequences, 0.3% of them were of rRNA gene origin. For the RNA sequences, the percentage of rRNA sequences varied between 58 and 76% with the exception of the surface sample collected at 5 m, which was dominated by eukaryotic rRNA sequences (.80%). The high count of eukaryotic rRNA sequences in the surface indicates that a further depletion of eukaryotic rRNAs supplementing the depletion of prokaryotic rRNAs would have been worthwhile. Nevertheless, also the depletion of prokaryotic rRNA still resulted in relatively high counts of bacterial rRNAs (7-55%), whereas the number of archaeal rRNAs was mostly negligible (2-11%). The rRNAs and rRNA genes were excluded from further analysis (except for Figure 3A), leaving 1,882,842 DNA and 421,528 RNA sequences. From this pool, 54% (DNA) and 53% (RNA) of the sequences could be identified as protein-coding (matching either the nonredundant database of NCBI or profile hidden Markov model scans of the ModEnzA EC groups or the Pfam protein families); the remainder could not be assigned. Of all protein-coding sequences, 82% of the DNA and 99% of the RNA sequences could be taxonomically identified (matched the non-redundant database of NCBI).

Hierarchical clustering of sequences using the taxonomic profiles
The taxonomic profiles of the metagenomes and the metatranscriptomes (the occurrence of the microbial taxa in the samples as a percentage of the total number of sequences having a BLASTxhit) as presented in Figure 3 were used for a hierarchical clustering with the PRIMER 6 program [102]. The samples were grouped into 6 categories according to the depths and a multivariate statistical test (ANOSIM) was used to determine if the groupings were distinct from each other in terms of their microbial communities. The relationships between the depth groups were visualized in a non-parametric Multidimensional Scaling plot using PRIMER 6.

Metabolic and taxonomic diversity measures
The EC activity matrix (with sample sizes equalized to the smallest sample) was exported from the FROMP pathway mapping tool [96] and the EC counts for each sample were used to calculate the inverse of Simpson's index (1/D) where D = S P i 2 and P i representing the proportional abundance of species i, and the Evenness E = (1/D)/S with S being the number of unique species. Similar calculations were also performed for the taxonomic assignments (as presented in Figure 3) from the BLASTx-searches normalized to total number of sequences having a BLASTx-hit.

Sequence recruitment onto reference genomes
The sequence data was recruited onto the reference genomes of the five most abundant organisms (as detected with BLASTsearches) SUP05, R. magnifica, V. okutanii, Sulfurovum and D. autotrophicum using the MUMmer program [103]. For SUP05, the draft genome (the metagenome with an ordered assembly of the contigs) as provided by Walsh and colleagues [41] was used and treated like the other genomes. The recruited sequences were re-assessed using a BLAST-search against the reference genomes. Sequences that hit more than one genome with a bit score difference of less than 5% between the first and second hits were discarded, giving rise to a non-overlapping set of sequences for each genome. These sequences were then recruited onto the genomes again to calculate the average coverage over nonoverlapping windows of 300 base pairs. The coverage of RNA sequences in every reference genome window was normalized by the total number of BLASTx-hits for that metatranscriptome and divided by the coverage of DNA sequences (which have also been normalized by the total number of BLASTx-hits in that metagenome) for the same window. This value, the expressionratio, was subsequently corrected for differences in sizes of the metatranscriptome and metagenome and selected regions of the reference genomes were plotted using customized R and PERL scripts.

Flow-cytometry cell counts
Samples for flow-cytometry were fixed with a final concentration of 1% paraformaldehyde and stored at 280uC until analysis. Total microbial cell counts were performed on a FACSCalibur flow cytometer (BD Biosciences). After 20 minutes staining of the samples with SybrGreen (Qiagen) at 4uC, cells were counted for 2 minutes or until a total count of 50,000 was reached. Sample flow rate was calibrated with standard beads (Trucount, BD Biosciences) and cell numbers were calculated via the time of measurement.

Chemical analysis and microsensor measurements
Our pump-CTD was equipped with a custom-built amperometric O 2 microsensor to obtain vertical profiles of dissolved O 2 (detection limit 0.5-1 mM). In addition, the recently developed self-calibrating Switchable Trace amount Oxygen (STOX) sensor was deployed, which allows high-accuracy O 2 measurements in near anoxic environments (detection limit ,50 nM during our measurements) [13,16]. After a minimum of ten minute sensor equilibration at a given sampling depth, at least five measuring cycles were used to calculate O 2 concentrations.
Water samples for nutrient analysis were taken with a depth resolution of 1 to 2 m. NH 4 + was measured fluorometrically [104] and NO 2 2 was analyzed spectrophotometrically [105] on board. Water samples for NO 3 2 and PO 4 32 were stored frozen until spectrophotometric determination [105] with an autoanalyzer (TRAACS 800, Bran & Lubbe). Detection limits for NH 4 + , NO 2 2 , NO 3 2 and PO 4 32 were 10, 10, 100 and 100 nmol l 21 , respectively. Dissolved N 2 O concentrations were determined on board in triplicates measurements using the GC headspace equilibration method as described elsewhere [106].
H 2 S concentrations were measured continuously on water sampled from the pump-CTD using a custom-built microsensor with a detection limit of ,0.5 mM [107]. Chemically determined H 2 S concentration (both H 2 S and HS 2 ; detection limit ,0.5 mM) on discrete water samples were used to calibrate the microsensor [108]. Although the sulfidic waters contained a composite of H 2 S and HS 2 , we use the term H 2 S throughout the manuscript to avoid unnecessary complexity.
Using a 125 m depth cut off and a grid resolution of 19 (bathymetrical data was obtained from the National Geophysical Data Center) along the ,200 km cruise track on the Peruvian shelf where H 2 S was detected in the water column, a total area of ,5500 km 2 was calculated to be affected by the sulfidic plume. The shelf contained mean H 2 S-maxima close to the bottom of the water column of 3.4 mM. H 2 S was extending vertically over the water column with an averaged depth of ,80 m, yielding ,440 km 3 of H 2 S-containing waters. Based on an average H 2 S concentration of 1.5 mM, we estimated a total content of ,2.2610 4 tons H 2 S within the plume.

Flux calculations
The density was calculated using the data processing program SeaSoft (Sea-Bird Electronics). The stability of the water column was expressed using the Brunt-Vä isälä frequency N, defined as: N 2 = 2(g/r)6(hr/hz) where g is the gravitational acceleration, r the water density and z the water depth. The density gradient was calculated over 4 m bins. The turbulent diffusivity Ez was calculated as described earlier [109] from the Brunt-Vä isälä frequency and the dissipation rate of turbulent kinetic energy e: Ez = ce/N 2 with the mixing coefficient c = 0.2. We applied a mean e of 1.85610 29 W kg 21 [110]. This value was measured for the open-ocean thermocline [110] and was applied in several rate diffusion models [111,112]. Vertical concentration gradients for O 2 , H 2 S, and NO 3 2 were calculated over 4 m bins. Fluxes of O 2 , H 2 S, and NO 3 2 at respective depths were calculated according to Fick's law: J i = Ez6(hC/hz).

Satellite images
Data of the sensors MODIS (Moderate Resolution Imaging Spectroradiometer) aboard the satellites Aqua and Terra (NASA) as well as MERIS (Medium Resolution Imaging Spectrometer) aboard the satellite Envisat (ESA) were implemented to study milky turquoise discolouration in waters off the Peruvian coast. Cloudy weather north of Pisco during most of our cruise made remote sensing difficult, but turquoise discolorations were observed off Lima on January 20-21 st ( Figure S1A) and 27-28 th ( Figure S1B). The estimation of the extent of the plumes requires data of higher spatial resolution; the algorithm for the identification of colloidal S 0 plumes is based on the high spectral resolution of MERIS data [57]. Since full resolution MERIS data is not available for this region we present a high resolution MODIS true colour image in our main figure ( Figure 1E).
However, the reflectance spectra derived from reduced resolution MERIS data ( Figure S1C), revealed that the turquoise plume southwest of Pisco conformed the criteria for S 0 in the identification algorithm distinguishable from optically similar coccolithophore blooms [113]. Differences in the shape and appearance of the colloidal S 0 plume on the MERIS image ( Figure  S1C) and the MODIS images ( Figure 1E) from May 7 th and 8 th demonstrate the temporal variability, which is also visible in changes in the cloud structure between the two images.

Rate measurements of nitrogen transformation processes
Rates of microbial nitrogen transformations were measured at three or four depths in 15 N-labeling experiments as described previously [114,115]. Briefly, nitrogen-loss via anammox and denitrification as well as dissimilatory NO 3 2 reduction to NO 2 2 and NH 4 + were measured in short-term incubation experiments amended with either 15 NO 3 , 15 NO 2 2 , 15 N 2 O or 15 NH 4 + (20, 10, 1 and 5 mmol l 21 , respectively; isotopes: Campro Scientific). Timeseries incubations were carried out in 12 ml Exetainers (Labco) and biological activity was stopped in one replicate Exetainer at each time interval (0, 6, 12, 24 and 48 h) by the addition of saturated mercuric chloride. Anammox and denitrification was measured as the production of 15 N-labeled N 2 in 15 NO 2 2 and 15 NO 3 2 , 15 NO 2 2 and 15 N 2 O incubations, respectively. The Nisotopic composition of N 2 gas produced in these experiments was determined by gas chromatography isotope-ratio mass spectrometry (GC/IRMS, Fisons VG Optima) [115]. NO 2 2 produced from 15 NO 3 2 and NH 4 + produced from 15 NO 2 2 was determined by GC/IRMS after conversion of NO 2 2 and NH 4 + to N 2 by sulfamic acid [114] and alkaline hypobromite [24], respectively. Production rates were calculated from the increase of 15 N-concentrations over time and only significant and linear productions of 15 N-species without a lag-phase were considered (t-tests, p,0.05; R 2 .0.8).
Rates are presented as net production rates corrected for the mole fractions of 14 N in the original substrate pools.

Rate measurements of carbon fixation
Triplicate incubations of 4.5 l seawater were set up with water from 5, 15, 30, 80 and 100 m. To each bottle 4.5 ml of 13 C bicarbonate solution (1 g 13 C bicarbonate in 50 ml water) was added and bottles were incubated in on-deck incubators shaded to 25% surface irradiance with blue lagoon light foil (Lee Filters) and continuously cooled with surface seawater (5 and 15 m) or incubated at 12uC in the dark (30, 80 and 100 m) for 24 hours. At the end of the incubation 1-2 l were filtered on precombusted (450uC for 6 hours) Whatman GF/F filters (as these filters have an average pore size of 0.7 mm and small microorganisms may have passed through, the calculated CO 2 fixation rates we present here have to be considered as minimal rates). The filters were oven dried (50uC for 24 hours) and stored for later analysis at room temperature. Filters were smoked overnight with 37% HCl to remove inorganic carbon retained on the filters, dried for 2 hours at 50uC and then analyzed in a CHN analyzer coupled to an isotope ratio monitoring mass spectrometer. The CO 2 fixation rate was calculated according to the enrichment of 13 C in the samples relative to unlabeled background values: C fix = (At% sample 2tAt% background )/(At% label 2At% background )6(POC/ time) where At% sample is the ratio of 13 C/ 12 C times 100 in the particulate organic carbon pool (POC), At% background is the same ratio in unlabeled POC and At% label is the final ratio of 13 C/ 12 C in the incubation bottle after label addition. The resulting ratio is multiplied with the concentration of POC and divided by the incubation time in days. Since we did not perform killed controls, we cannot exclude or estimate the contribution of anapleurotic carbon fixation in our samples. We averaged the CO 2 fixation over the predicted predominantly photic (0-20 m, ,14.4 mmol C l 21 d 21 ) and aphotic (20-100 m, ,1.2 mmol C l 21 d 21 , dark incubations) zones by multiplying the rates with the respective water depths. A photic zone CO 2 fixation of ,288 mmol C m 22 d 21 and a lightindependent CO 2 fixation of ,96 mmol C m 22 d 21 was estimated.
When compared to the overall mean CO 2 fixation rate of the Humboldt Current System off Peru (2.18 g C m 22 d 21 or 182 mmol C m 22 d 21 [2] and 3.5 g C m 22 d 21 or 292 mmol C m 22 d 21 [4]), our measured dark CO 2 fixation (over an 80 m deep aphotic zone) contributed 33-53% of the total CO 2 fixation. Extrapolating the dark CO 2 fixation rates over the entire sulfidic plume (,5500 km 2 ), we calculated a total CO 2 fixation of ,5.3610 8 mol C d 21 or ,6.3610 3 tons C d 21 . This CO 2 fixation estimate would contribute 1.2% of the total primary production of the Humboldt Current System off the Peruvian coast as presented by Carr, 2002 (,2610 8 tons C y 21 or ,5.5610 5 tons C d 21 ). The average CO 2 fixation rates per cell were calculated from total microbial cell counts as obtained from flow-cytometry.

Accession numbers of sequence data
Metagenomic and metatranscriptomic sequences have been deposited in the metagenomics analysis server (MG-RAST) under accession numbers 4460677. 3  Vertical distribution of sequences recruited onto the genomes of (A) Sulfurovum sp. NBC37-1 and (B) Desulfobacterium autotrophicum HRM2. Shown are selected genes encoding for enzymes involved in oxygen-(blue), sulfur-(yellow), nitrogen-(red), carbon-(green) and hydrogen-metabolism (purple) in the corresponding order of the genomes. The y-axis depicts the log of the expressionratio, a measure for the selective enrichment of transcripts over the corresponding gene, normalized to the total pool of protein-coding sequences. A list of the start and end position of each gene and the full name of the corresponding enzyme are shown in Table S3. (TIFF) Figure S5 Vertical distribution of the most abundant functional assignments. Shown are the top five most abundant EC numbers in percent of all protein-coding sequences in the DNA and RNA datasets; ordered descending according to the DNA counts and supplemented with the remainder of the top five EC numbers from the RNA dataset if not already present in the DNA dataset. Please note that the data presented here is based only on EC number-and Pfam-assignments; BLAST-hits are not included. (PDF) Table S1 Sequencing statistics. a as obtained with Cd-hit. b as obtained by BLASTn-searches against the SILVA database. c as obtained by BLASTx-searches against the non-redundant database of NCBI and by scans with profile hidden Markov models of the ModEnzA EC groups and of the Pfam protein families. d average. (DOC) Table S2 Metabolic and taxonomic evenness and diversity in all protein-coding sequences. A collection of all EC number-and Pfam-assignments was used to determine the metabolic diversity, while the taxonomic diversity was calculated using all hits from BLASTx-searches. Shown are the evenness and the diversity (inverse of the Simpson's index) for both the metagenomic and metatranscriptomic datasets.