Study Site

The oligotrophic Lake Stechlin is located in the Brandenburg-Mecklenburg Lake District, Germany (53°10’ N, 13°02’ E). Field permit was granted to the Leibniz-Institute of Freshwater Ecology and Inland Fisheries by the Stechlin Park nature authorities on a permanent basis. Lake Stechlin is a hard water lake which encloses an area of 4.25 km² with a maximum and mean depth of 68 m and 22.8 m, respectively. Average total phosphorus concentration is on average 13.1 μg L^-1 during summer time [16]. The entire water column remains oxic throughout the year, but in the sediments dissolved oxygen is depleted in the uppermost layers. The lake has a relatively high sulfate concentration of 200 μmol L^-1 in the water column. In the sediment, the upper 7 cm is depleted in sulfate and is characterized by elevated porewater sulfide concentrations [17]. Dumont et al. [18] found relatively low methane uptake rates under in situ conditions. Methane production primarily occurs in deeper sediment layers (20–25 cm), but could potentially take place in surface layers as well [19]. Denitrification is relatively low in the littoral sediments, but higher at the lake’s center [20]. We used profundal sediments for the experiments to obtain maximum microbial activity. The sediments at the lake’s centre were most promising in this regard due to their fine texture and relatively high organic matter content.

Experimental design

In June 2010, a 40 cm long, intact sediment core with an overlying water column of 20 cm was taken from the deepest point of the lake with a gravity corer (UWITEC, Mondsee, Austria). The core was hermetically sealed and mounted as shown in Fig 1A and 1B (without the electrochemical sensors). The overlying water was constantly circulated by a peristaltic pump (Abimed) to prevent chemical stratification. The core was sealed with airtight caps to exclude external oxygen. Incubation experiments were performed at room temperature (22–25°C) rather than the in situ temperature (6–7°C) so that changes in microbial activities would be more readily observable within the experimental period. Likewise, we added a high initial concentration of potassium nitrate (~3 mmol L^-1) to the oxic overlaying water to increase the redox buffer capacity. Goal of the prolonged presence of nitrate was to delay the subsequent processes in a way that rendered optimum timing of our subsampling. In the course of the experiment we repeatedly added oxygen (air bubbling) and/or potassium nitrate to check for microbial responses after the depletion of the particular electron acceptor.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design without electrodes.

(A) Valve settings during circulation via pumping to prevent a formation of diffusive boundary layer above the sediment and to ensure an equal distribution of solutes in the water column; (B) Valve settings during resuspension /sampling mode with pump off.

https://doi.org/10.1371/journal.pone.0143428.g001

Chemical measurements

In order to follow the hydrochemical changes in the overlaying water in real time, dissolved oxygen (DO), pH, oxidation–reduction potential (ORP) and hydrogen sulfide were monitored online as previously described [21]. Nitrate was measured by an ion-selective nitrate electrode (Mettler-Toledo, Gießen, Germany). Dissolved methane in the overlying water was measured daily using a modified version of the head space technique [22]. Two 1 ml subsamples were transferred into 10 ml crimp vials prefilled with N₂. After sealing the vials, dissolved CH₄ was extracted into the gas phase by vigorous shaking and, thereafter, analysed on a gas chromatograph with a flame ionization detector (Shimadzu, Duisburg, Germany).

Microbiological analysis

RNA extraction and cDNA synthesis

RNA for studying community composition of active microorganisms was extracted from samples collected in the experiment. We used chloroform-phenol-isoamyl alcohol and zirconium beads following the protocol of Nercessian et al. [23], except that the extraction buffer was prepared with 240 mM instead of 200 mM potassium phosphate buffer. The DNA- RNA-pellet was dissolved in 35 μL RNase-free water and stored at -80°C. DNA was digested twice with Turbo-DNA free kit (Ambion) following the manufacturer’s instructions, except that 2.5 μL instead of 2 μL of the reaction buffer was used. Successful removal of DNA was checked by PCR using the universal primer pair 341f/ 907r (S1 Table). Approximately 100 ng of total RNA was used for cDNA synthesis via the Array Script kit (Ambion). However, we were unable to produce cDNA from some of the samples, especially for sample points between 250–400 h and 500–600 h.

Polymerase chain reactions

The reaction mixtures for PCR amplifications contained 5–20 ng μL^-1 cDNA, 0.4–0.8 pmol μL^-1 of each primer (S1 Table), 250 μmol L^-1 of each dNTP, 3 mmol L^-1 MgCl, 2.5 μL of 10 × PCR buffer, and 0.5 U Biotaq Red DNA polymerase (Bioline) in a total volume of 50 μL. In addition to the universal Bacteria and Archaea primers, we used a set of primers suitable for amplifying genes of some key enzymes in elemental cycling. Unfortunately, the partial sequence of the ammonium monooxygenase gene (amoA) from bacteria and archaea could not be successfully amplified in a PCR approach. All successfully amplified genes and their functions are given in Table 1. Primer sequences and PCR conditions are provided in S1 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Applied PCR approaches and the function of their target genes.

https://doi.org/10.1371/journal.pone.0143428.t001

Data analysis

Transition phase 1: From oxygen to nitrate (0–130 h)

The initial overlying water was oxic with an ORP of 150 mV (Fig 2). After four hours of incubation we added the nitrate redox buffer, which raised the ORP to 380 mV. The initial 150 μM of dissolved oxygen (DO, not shown in Fig 2) in the overlying water was depleted within the first 10 h after the system was sealed off from external air. At the SWI, the depletion of DO triggered the first transition phase from oxygen as the electron acceptor to nitrate. This phase lasted for about 130 h and was accompanied by a gradual ORP decrease to -40 mV. DGGE 16S rRNA fingerprint analysis displayed a shift in the microbial community composition after 50 h: two prominent bands disappeared during the transition from oxic to anoxic conditions (S1 Fig). NMDS analysis showed a high variance within the first transition phase and no clear selection from the second transition phase (Fig 3). ANOSIM analysis showed significant differences between the clusters (p = 0.001, r = 0.558). Envfit analysis suggests methane and nitrate, but not redox potential, as significant triggers (p = 0.002 and 0.007, see S2 Table). The main environmental influences on DGGE patterns of “active” microorganisms and expressed functional genes must be interpreted carefully due to covariance of various chemical parameters in the sediments. DGGE analysis of active Archaea (16S rRNA), however, did not show any prominent change (S2 Fig). Expression patterns of the nirS gene for denitrification significantly changed after oxygen depletion and after 32 h when nitrate concentrations dropped below 2.0 mmol L^-1 and ORP was below 200 mV (S3A Fig). This was consistent with a preliminary experiment demonstrating rapid shifts in nirS expression in response to nitrate addition within one hour (Cluster analysis; S4 Fig). The expression pattern of nirS displayed a high diversity while band diversity of nirK was much lower (S3B and S5 Figs). Sulfate reducers and methanogenic archaea were detectable under oxic and denitrifying conditions. Although the sulfate reducing community (represented by dsrB expression patterns) revealed significant shifts between time points, no clear trend was observed during the transition from oxygen to nitrate as electron acceptor (Fig 4, S3A Fig). After the depletion of DO, dissolved methane increased from 0.02 to 0.1 mmol L^-1 within 48 hours, and remained at this level until reoxidation of the overlying water (hour 130). Active methanogenic archaea could be detected with 16S rRNA as well as mcrA specific primers at all times. DGGE pattern of archaeal 16S rRNA changed significantly due to ORP (p = 0.001, see S1 Table) but only in non-dominant bands (S2 Fig). McrA DGGE pattern was not influenced by the measured environmental parameters (oxygen: p = 0.04, but did not remain significant after Bonferroni-Holm correction, S2 Table). This indicates that ORP did not play a crucial role in shaping the community of active Archaea and particularly methanogens at the SWI. In contrast, pmoA expression (representing active methanotrophy) was detected repeatedly during the experiment without using a nested PCR approach as long as oxygen and/or nitrate was present. This suggests that aerobic oxidation of upward diffusing methane prevented methane accumulation in the overlying water. As long as nitrate was present in the overlying water, anaerobic oxidation of methane with nitrate (AOM) limited CH₄ increase to 0.1 mmol L^-1. Spearman's rank correlation analysis showed that nitrate and methane values are negatively related (r = -0.53, p < 0.001). Similar trends in chemical parameters were observed in a preliminary experiment in 2007 (S6 Fig). Sequence analysis of active methanotrophs revealed no changes in community composition, despite the absence of oxygen.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Hydrochemical changes in the overlying water of an incubated sediment core during redox succession.

The time line above indicates the different microbial phases within the experiment and highlights the preferred electron acceptors within these phases. Ox: Oxygen consumption; DN: Denitrification; AOM: anaerobic oxidation of methane; MA: Methane Accumulation; SR: Sulfate Reduction; SO: Sulfide oxidation.

https://doi.org/10.1371/journal.pone.0143428.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. NMDS plot (stress 0.13) of 16S rRNA bacterial DGGE banding patterns.

Arrows indicate the direction triggered by methane and nitrate as the strongest analyzed environmental factors. Ellipses indicate a confidence interval of 95% and ANOSIM analysis was significant (p = 0.001, r = 0.558). Colors indicate the six transitions phases (TP): red = TP 1 (Transition from oxygen to nitrate as electron acceptor), green = TP 2 (Short oxic pulse), yellow = TP 3 (nitrate depletion and sulfide development), light blue = TP 4 (Nitrate additions under sulfidic condition), dark blue = sulfidic phase, purple = oxic phase.

https://doi.org/10.1371/journal.pone.0143428.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Development of dsrB and nirS DGGE band diversity in relation to ORP.

https://doi.org/10.1371/journal.pone.0143428.g004

Transition phase 2: Reaction to DO pulse (130–230 h)

After 130 h of incubation, we added 165 μmol L^-1 oxygen to the overlying water. As a result ORP increased rapidly to 300 mV and dissolved methane dropped to 0.02 mmol L^-1. The occurrence of oxic processes was not reflected by changes in DGGE banding patterns of the active bacterial community. NMDS analysis, however, revealed a second cluster corresponding to the second transition phase including the samples taken when nitrate was below the detection limit (130–230 h; Fig 3). Magnetospirillum appeared to be the predominant species with up to three distinct bands (185–240 h; S1 Fig). We found no clear reaction of the active microbial community to the oxygen pulse, which lasted only for about 15 h when ORP dropped from ca. 300 mV back to its previous level of -40 mV. Methane increased again from 0.01 to 0.1 mmol L^-1. Two previously dominant 16S rRNA bands from the first oxic phase vanished after oxygen depletion and did not reoccur despite subsequent oxygen additions (S1 Fig, red squares). Nitrate remained detectable only up to 200 h, but ORP suggests that it was still present below detection (<0.01 mmol L^-1) until 230 h. The nirS expression pattern (denitrification) changed after 146 h in the absence of nitrate, indicating changes within the active denitrifying community. Expression patterns of mcrA (methanogenesis) and nirK (denitrification) genes in relation to the environmental parameters did not change significantly throughout the whole experiment (see S2 Table). Interestingly, there seems to be a slight effect on mcrA band diversity, since band number was highest at transition phases, which was not the case for nirK (S5 Fig). AprA (sulfide oxidation and sulfate reduction) and pmoA (methanotrophy) genes exhibited changes which could be related to fluctuations in ORP, (p = 0.011), methane (p = 0.001) and pH (p = 0.006). For pmoA, 31 of the 37 sequences were affiliated with methanotrophic type Ib Methylococcaceae occurring at different heights on the DGGE gel. Only four of the identified pmoA sequences in this study belonged to the type Ia methanotrophs (Fig 5). Only one sequence was classified as methanotrophic bacteria type II (Methylocystis) which was detected only under oxic conditions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Bar chart of analyzed functional genes and their phylogenetic classification.

(A) Phylum-level classification of functional gene sequences; (B) Phylogenetic classification of the sequences to their closest related family. Colors correspond to encountered phyla, while patterns indicate families.

https://doi.org/10.1371/journal.pone.0143428.g005

Transition phase 3: Depletion of nitrate and first sulfidic phase (230–400 h)

With the depletion of nitrate at 230 h, the ORP signal dropped rapidly from 0 to -250 mV and then gradually to -480 mV (Fig 2). From that time point onwards sulfide concentration increased in the overlying water. Interestingly, dissolved methane that was absent under oxic condition and present at low levels before the depletion of nitrate, increased in parallel to sulfide concentration. With the ORP dropping below -200 mV, bacterial community composition showed a third cluster in the NMDS (Fig 3). During this phase, Magnetospirillum appeared to be the predominant bacterial phylotype occurring in the 16S rRNA DGGE gel (S2 Fig). Sulfate reducers (dsrB expression patterns) showed a strong increase in diversity when the ORP dropped from -40 to -500 mV, which was not the case for nirS (Fig 4). The main DGGE bands in nirK expression patterns vanished between 210–230 h, and three additional bands occurred exclusively during the following transition phase. After the transition phase, nirK expression patterns were similar to those before the transition, although they seemed to be less diverse. At 400 h both methane and total sulfide reached concentrations of 0.5–0.6 mmol L^-1.

Transition phase 4: Nitrate addition under sulfidic conditions (400–570 h)

Nitrate was added a second time at 400 h to a final concentration of 1 mmol L^-1. As a consequence, sulfide was oxidized completely within 30 h, and methane concentrations declined from 0.55 mmol L^-1 at 400 h to 0.4 mmol L^-1 at 500 h when nitrate concentration dropped below the detection limit (Fig 2). Interestingly, ORP increased in the presence of nitrate to the same values (~ -40 mV) as during the previous nitrate addition in the absence of DO. The observed hydrochemical changes led to a physiological response as indicated by the fourth, distinctly different cluster in the NMDS analysis based on 16S rRNA DGGE banding patterns (Fig 3). The dsrB expression pattern changed with nitrate, as indicated by a sudden decrease in DGGE band diversity (Fig 4, S3 Fig). Denitrifiers expressing nirS did not show major changes in DGGE banding patterns upon the second nitrate addition. DGGE bands that were prominent after the first nitrate addition did not reoccur. In line with the decrease in methane, pmoA could be amplified again without using a nested PCR approach.

Interestingly, aprA genes affiliated to Hydrogenophilaceae were expressed when sulfide was present, but were not detectable when nitrate concentrations were high and sulfide not detectable. After nitrate depletion (570 h), aprA genes of the organism were expressed again. In contrast, the prominently expressed aprA genes associated with Deltaproteobacteria were always detectable between 230–640 h, i.e. the first depletion of nitrate until the reintroduction of DO.

Transition phase 5: Second sulfidic phase (570–630 h)

Right after the depletion of nitrate at ~570 h, sulfide and methane concentrations increased to 0.03 and 0.58 mmo L^-1, respectively, while ORP dropped back to -500 mV (Fig 2). Analysis of a single sample taken during the prominent ORP drop with increasing sulfide concentrations confirmed our results from the first sulfate reduction phase. At that time, expressed dsrB gene diversity strongly increased during the transition from oxidized nitrogen species to sulfate as the electron acceptor (Fig 4, S3A Fig). In contrast, expressed aprA gene pattern did not change during this transition phase.

Transition phase 6: third oxic phase (> 630 h)

At T = 630 h, the overlying water was turned oxic again, leading to the rapid oxidation of methane and sulfide. Apparently this drastic redox change did not result in measureable effects on the expression of functional genes or the composition of active microorganisms (Fig 2, S1 Fig). These results match observations made in the aftermath of the DO addition at 130 h.

Composition of the active microbial community

Most of the active bacteria community belonged to Betaproteobacteria (56%), while only a few sequences belonged to Alphaproteobacteria (16%), Bacteriodetes (6%), Cyanobacteria (9%) and Epsilonproteobacteria (3%) (S7A Fig). Several DGGE bands of bacteria that were present over the course of the experiment were identified as members of Caulobacteraceae and Rhodospirillaceae.

16S rRNA DGGE analysis did not reveal any Delta- or Gammaproteobacteria affiliated with aprA, dsrB and pmoA. Phylogenetic classification of 7 aprA and 32 dsrB sequences indicate that these sequences were affiliated mainly with Deltaproteobacteria such as Desulfobacteraceae and Desulfobulbaceae (Fig 5). Only a few sequences were classified as Firmicutes. Over 80% of the retrieved expressed pmoA sequences were affiliated with methanotrophic type Ib Methylococcaceae (Fig 5). However, these phyla could not be found by using the 16S rRNA analysis.

31 nirS sequences were analyzed, whereby 19% belonged to Betaproteobacteria, 42% to Alphaproteobacteria, and 29% could only be classified as Proteobacteria. Achieved nirK sequences were predominantly classified as Alphaproteobacteria.

All sequenced active Archaea belonged to Euryarchaeota classified as Methanospirillaceae, Methanoregulaceae, Methanosaetaceae and Thermoplasmatales belonging to the CCA47 group (S7B Fig). Dominant sequences were affiliated with Methanospirillaceae (47%) and Methanoregulaceae (40%). However, when analyzing the expression of the mcrA gene, none was affiliated with the Methanospirillaceae, although this family dominated the active Archaea community. In contrast, two prominent expressed mcrA sequences were distantly related to Methanoculleus.

Limitations of DGGE analysis

To avoid methodological biases caused by extracellular DNA in sediments, we focused on the 16S rRNA representing the amount of ribosomes in the cells, and mRNA as an indicator of gene expression. As for all PCR based methods, our results may have been biased by the presence of multiple copies of the target gene, potential primer biases forming multiple or artificial bands on the DGGE gel, or simply a failure of primer binding to potential targets [28–30]. The 16S rRNA analysis via DGGE only gave a partial picture of the active microbial community because organisms of less than 1% of the community may not produce visible DGGE bands [31]. Comparison of 16S rRNA data with representatives of functional gene assays showed only minor similarities, which might be a result of insensitivity of the DGGE system. The lack of gammaproteobacterial and deltaproteobacterial sequences in 16S rRNA analysis was also unexpected since these organisms are common in lake sediments and also in Lake Stechlin ([32,33], Frindte et al., unpublished results). It is possible that the 341f and 907r/803r system did not amplify these sequences, but in silico analysis in TestPrime (http://www.arb-silva.de/browser/ssu-122/silva-ref-nr/testprime) did not support this assumption. It is likely that the DGGE fingerprint analysis was too insensitive and/or that diversity of individual taxonomic units was relatively high, so that no visible bands occurred on the DGGE gels. Consequently, 16S rRNA analysis via DGGE might have underestimated the diversity and thus potential changes due to changes in environmental conditions. New high throughput sequencing techniques have the potential to overcome such limitations, but require enormous resources especially if the aim is to analyze the expression of different genes over time.

Redox transitions and functional gene activity

Denitrification became the dominant process following the depletion of oxygen with a characteristic ORP of -40 mV at three occasions. What differed between those events was the response time of the system to establish a stable Pt-electrode reading. The first ORP drop from 360 to -42 mV took 100 h, whereas after the second DO pulse, the associated ORP decreased from 260 to -40 mV in only 10 h. Apparently, other oxidized electron acceptors like manganese, which is present in Lake Stechlin, buffered the system during the first ORP decline [34]. A similar redox buffer capacity has been suggested previously, e.g., for tropical soils [35,36]. The DGGE banding pattern present under the first oxic phase did not reoccur at the end of the experiment (e.g., bacterial 16S rRNA, nirS and dsrB) when the system returned to being oxic; thus, it seems that the microbial community did not recover from the intervening anoxic condition. Most likely, the recovery time of the microbial community in the aftermath of a prolonged anoxic phase is the result of high resilience [37] and only a small fraction possesses the ability to react more directly on environmental changes [38]. Since we did not analyze the community composition based on DNA, it is likely that we have overlooked many microorganisms present in a dormant stage in the sediment. 16S rRNA analysis showed that the active bacterial community changed little after 400 hours. Crump et al. [39] identified sulfide as the main driver of the microbial community composition in seasonally anoxic estuarine bottom waters, which also might be true for our simulation experiment. In contrast, expression patterns of some functional genes varied throughout this period (e.g., dsrB and aprA), indicating a much higher sensitivity towards temporal fluctuations in environmental conditions. However, after re-oxidation, prominent members of the active microbial community under oxic conditions did not reoccur, indicating the disappearance of organisms that are not well adapted or may need more than a few days to recover as has been shown for seasonally anoxic waters [15,39].

Therefore, it is not surprising that induced redox transitions revealed the strongest changes in functional expression patterns of sulfate reducers (dsrB, Fig 4) and Cd1-dependent nitrite reducers (nirS), especially during transition phase 3 when the system shifted from nitrate rich to sulfidic conditions. The high diversity in functional gene expression pattern of sulfate reducers (dsrB) during the transition from nitrate to sulfate reduction indicates that this critical redox transition stimulates gene expression and hence activity of sulfate reducers. In contrast, nirS was expressed while oxygen was still present indicating the simultaneous occurrence of oxic respiration and denitrification processes [40,41]. On the other hand, addition of nitrate to the system greatly reduced the number of expressed dsrB genes suggesting that nitrate addition under anoxic conditions results in reduced dsrB gene expression until nitrate depletion. Oxygen did not inhibit dsrB gene expression (as observed for nirS) indicating that NO_x concentrations rather than the redox potential itself was responsible for the observed redox-sensitivity in dsrB gene expression [17,41].

Fluctuation in redox conditions and methane cycling

Concentration of dissolved methane in the overlying water is regulated by the interplay of physical processes (diffusion and ebullition) and methanotrophy. Methanogenesis, the source of biogenic methane, is restricted to the deeper sediment layers that are deplete in sulfate [19,42]. The resulting concentration gradient of dissolved methane between the sediment pore water and overlying water leads to diffusion of methane into the overlying water. In the presence of oxygen, aerobic methanotrophic bacteria efficiently prevent methane enrichment at the SWI, as shown earlier by microprofile measurements in lake sediments [43,44], and is also consistent with our methane measurements during the three oxic phases (Fig 2 at T < 10 h, T = 120–140 h, and T > 620 h). The increase in methane in the overlying water after DO depletion, however, was not uniform, and the resulting accumulation pattern differed from that of sulfide. The presence of nitrate seemed to play a critical role by blocking any release of sulfide at the SWI and by controlling methane release rates. Apparently, the plateau of ~ 0.1 mmol L^-1 of dissolved methane reached at T = 50–120 h and at T = 190–220 h was the result of anaerobic oxidation of methane (AOM) using nitrate as the electron acceptor [45–47]. The finding that methane increased rapidly at a rate of 50 μmol L^-1 d^-1 following the depletion of nitrate at T = 220 h (Fig 2) strongly indicates the presence of AOM. In line with this notion, the expressed pmoA sequences mainly belonged to Methylococcaceae and, surprisingly, they did not vary much during the transition from oxic to anoxic conditions as long as nitrate was present. In our experiment, the presence of oxygen might be an explanation for effective aerobic methane oxidation at the beginning of the experiment, but this is unlikely when ORP decreased to -40 mV. In this case, AOM using oxidized nitrogen species as electron acceptors offers the most likely explanation [45,46]. From the parallel increase of sulfide and methane at T > 220 h (Fig 2), we may conclude that methanotrophs in our experiment were incapable of AOM with sulfate, a process common in marine sediments [48]. Apparently, the sediments in the oligotrophic Lake Stechlin harbour the potential for AOM via NO_x compounds despite the very low ambient nitrate and nitrite concentrations [46,47]. The addition of nitrate at T = 400 h caused a fast depletion of sulfide. We calculated an oxidation rate of sulfide with nitrate of 50 μmol L^-1 h^-1, which is equivalent to that with oxygen. Methane concentration, on the other hand, showed a gradual decline with nitrate as the electron acceptor at a rate of 1.3 μmol L^-1 h^-1 until nitrate depletion. Taking into consideration the measured methane release rate of 2.5 μmol L^-1 h^-1, methane oxidation rates with nitrate were 3.8 μmol L^-1 h^-1 while under oxic conditions they reached ca. 12.5 μmol L^-1 h^-1 –still five times lower than for sulfide. Thus, it seems likely that sulfide oxidation with nitrate is energetically more efficient than AOM with nitrate. This is consistent with the findings by Ettwig et al. [46], who posited that AOM with NO_x compounds is mainly nitrite dependent, and needs the denitrification steps to procure nitrite for the intracellular oxygen production.